1. A deep learning-based human activity recognition system (100), the system (100) comprising: a user device (102) integrated a plurality of cameras (104) and configured to capture high-resolution images and videos for real-time monitoring, analysing movements and behaviours from a user through a user interface (106); a plurality of sensors (108) positioned within the user device (102) and configured to sense motion, location, orientation, and environmental conditions of a user; a communication network (114) configured to establish a communication link for data transmission within the system (100); a processing unit (116) connected to the user device (102), the plurality of cameras (104), and the plurality of sensors (108) via the communication network (114), the processing unit (116) configured to process real-time data for human activity recognition, wherein the processing unit (116) further comprises: a data input module (118) configured to receive real-time data from the plurality of cameras (104), and the plurality of sensors (108) positioned within the user device (102); a pre-processing module (120) configured to clean, normalize and reduce noise in the received data to enhance the data quality; a feature extraction module (124) configured to extract relevant features from the pre-processed data including motion patterns, gestures, postures, and other relevant feature parameters; a human activity detection module (126) configured to analyse the extracted features and determine the activities performed by the user; a human activity classification module (128) configured to recognize and classify the detected human activity into predefined categories including presence of motion, gestures, change in postures, object interactions, cyclic actions and multi-user activities; an output module (134) configured to transmit the extracted features, detected human activity and classified human activity to the user device (102).

2. The system (100) as claimed in claim 1, wherein the system (100) further comprises a cloud database (136) configured to store and managed data associated with human activity recognition, thereby enabling remote access and improved classification accuracy.

3. The system (100) as claimed in claim 1, wherein the plurality of sensors (108) comprises an accelerometers (110) and a gyroscopes (112) configured to sense motion, orientation, and environmental conditions of a user for real-time human activity recognition.

4. The system (100) as claimed in claim 1, wherein the pre-processing module (120) incorporates a dynamic sensor reliability weighting engine (122) that evaluates the real-time performance of the plurality of sensors (108) and the plurality of cameras (104) configured to scale the corresponding sensor data according to its reliability, so that the weighted sensor data is provided as input to a deep learning model, thereby improving the accuracy and robustness of activity recognition and classification.

5. The system (100) as claimed in claim 1, wherein the human activity detection module (126) configured to determine and track the activities of multiple users within the same environment, thereby enabling accurate recognition of overlapping human activities including but not limited to simultaneous gestures, interacting tasks, cyclic motions, and concurrent movement patterns.

6. The system (100) as claimed in claim 1, wherein the human activity classification module (128) applies a trained deep learning model selected from artificial neural network, convolutional neural network, recurrent neural network, and a hybrid architecture to classify user activities in real-time.

7. The system (100) as claimed in claim 1, wherein the processing unit (116) further comprises an alert generation module (130) configured to generate a real-time alerts upon detection of unsafe activities including but not limited to sudden falls, prolonged inactivity, abnormal movement patterns and unauthorised access into restricted areas.

8. The system (100) as claimed in claim 1, wherein the processing unit (116) further comprises a training and testing module (132) configured to split the data into training and testing data sets to train the system (100), evaluates its performance, improves the system (100) stability and interpretability.

9. A method (200) for deep learning-based human activity recognition system (100), the method (200) comprising: capturing high-resolution images and videos for real-time monitoring, analysing movements and behaviours from a user through a user interface (106) via the user device (102); sensing motion, location, orientation, and environmental conditions of a user via the plurality of sensors (108); establishing a communication link for data transmission within the system (100) via the communication network (114); processing real-time data for human activity recognition via the processing unit (116), comprising several modules; receiving real-time data from the plurality of cameras (104), and the plurality of sensors (106) positioned within the user device (102) via the data input module (118); cleaning, normalize and reduce noise in the received data to enhance the data quality via the pre-processing module (120); extracting relevant features from the pre-processed data including motion patterns, gestures, postures, and other relevant feature parameters via the feature extraction module (124); analysing the extracted features and determine the activities performed by the user via the human activity detection module (126); recognizing and classify the detected human activity into predefined categories including presence of motion, gestures, change in postures, object interactions, cyclic actions and multi-user activities via the human activity classification module (128); generating real-time alerts upon detection of unsafe activities including but not limited to sudden falls, prolonged inactivity, abnormal movement patterns and unauthorised access into restricted areas via the alert generation module (130); splitting the data into training and testing data sets to train the system (100), evaluates its performance, improves the system (100) stability and interpretability via the training and testing module (132); transmitting the extracted features, detected human activity and classified human activity to the user device (102) via the output module (134); displaying extracted feature, detected human activity, classified human activity on the user interface (106) of the user device (102).

10. The system (100) as claimed in claim 1, wherein the Human Activity Detection Module (126) configured to measure the duration of each detected activity and provide this information to the human activity classification module (128) to improve the accuracy of classifying activities exhibiting time-dependent patterns, such as repeated motions, prolonged inactivity, and sequential tasks.

Description:FIELD OF DISCLOSURE
[0001] The present disclosure generally relates to deep learning techniques for human activity recognition in Internet of Things-enabled smart environment, more specifically, relates to a deep learning-based human activity recognition system.
BACKGROUND OF THE DISCLOSURE
[0002] An adaptive, edge-deployable human activity recognition system using a lightweight artificial neural network architecture. The system processes multi-sensor Internet of Things data in real time, dynamically adjusts sensor reliability weights, and supports energy-efficient, privacy-preserving activity classification. The system also enables continuous behavioural adaptation and integrated into existing smart home, healthcare, and industrial Internet of Things environments without extensive hardware modification.
[0003] Traditional human activity recognition systems in Internet of Things environments rely primarily on centralized cloud processing, conventional machine learning models, and resource-intensive deep learning architectures such as Convolutional Neural Networks and Recurrent Neural Networks, which analyse raw sensor data and video streams to identify human activities. These systems often require high computational power and memory, leading to limited deployment on low-power edge devices, increased latency, high bandwidth consumption, and potential privacy with security risks due to transmission of sensitive data. Furthermore, conventional approaches typically struggle to handle noisy and missing sensor data, lack adaptability to individual user behaviours and exhibit limited fault tolerance when sensors produce unreliable readings. In addition, many existing systems are unable to accurately recognize overlapping activities performed by multiple users, and they often fail to provide real-time responses in crowded environments. Traditional Human Activity Recognition solutions also generally lack personalization capabilities, making it difficult to account for variations in individual movement patterns, habits, and environmental conditions. Moreover, integration of heterogeneous sensors and devices is often cumbersome due to incompatible data formats and communication protocols, and many systems are difficult to retrofit into existing smart home, healthcare, and industrial Internet of Things infrastructures without significant hardware modifications and additional cost. Conventional systems are also generally incapable of performing early anomaly detection, limiting their usefulness in preventive healthcare, safety monitoring, and industrial process optimization. Additionally, most traditional solutions unable to include mechanisms for energy-efficient operation and adaptive resource management, resulting in excessive power consumption and reduced operational longevity of battery-powered Internet of Things devices. Traditional systems often rely on static models that require frequent retraining with large labelled datasets, which is time-consuming, expensive, and impractical in dynamic real-world environments. Furthermore, these systems typically offer limited interpretability, making it difficult for users to understand the rationale behind activity recognition decisions. Finally, conventional approaches generally unable to provide mechanisms for scalable deployment across multiple devices, leading to high maintenance overhead and reduced applicability in large-scale smart environments.
[0004] The present invention solves the problem of the prior art by providing an adaptive, edge-deployable human activity recognition system that processes multi-sensor Internet of Things data in real time using a lightweight artificial neural network architecture. The invention including reduced computational and memory requirements that enable deployment on low-power edge devices, enhanced energy efficiency, and minimized data transmission to protect user privacy and improve security. Furthermore, the invention is capable of dynamically handling noisy, missing, and unreliable sensor data through adaptive sensor reliability weighting, thereby improving fault tolerance and recognition accuracy. The invention supports personalization by adapting to individual user behaviours and environmental conditions, accurately recognizes overlapping in multi-user scenarios, and enables real-time responses in dynamic environments. In addition, the invention facilitates integration with heterogeneous sensors and devices, allows retrofit deployment in existing smart home, healthcare, and industrial Internet of Things infrastructures, and provides predictive activity analysis for preventive healthcare, safety monitoring, and industrial process optimization. The invention also incorporates mechanisms for scalable deployment, continuous learning, and resource-adaptive operation, resulting in longer battery life for edge devices and reduced operational costs.
SUMMARY OF THE DISCLOSURE
[0005] The following is a summary description of illustrative embodiments of the invention. It is provided as a preface to assist those skilled in the art to more rapidly assimilate the detailed design discussion which ensues and is not intended in any way to limit the scope of the claims which are appended hereto in order to particularly point out the invention.
[0006] According to illustrative embodiments, the present disclosure focuses on a deep learning-based human activity recognition system which overcomes the above-mentioned disadvantages or provide the users with a useful or commercial choice.
[0007] An objective of the present disclosure is to enable edge-deployable activity recognition, reducing reliance on centralized cloud processing.
[0008] Another objective of the present disclosure is to design a lightweight artificial neural network architecture that minimizes computational and memory requirements.
[0009] Another objective of the present disclosure is to achieve energy-efficient operation, allowing deployment on low-power Internet of Things devices and extending battery life.
[0010] Another objective of the present disclosure is to implement privacy-preserving mechanisms that minimize data transmission and protect sensitive user information.
[0011] Another objective of the present disclosure is to integrate multimodal sensor data, including accelerometers, gyroscopes, cameras, and environmental sensors, to improve the accuracy and robustness of human activity recognition.
[0012] Another objective of the present disclosure is to automatically extract features from raw sensor data using deep learning techniques, eliminating the need for manual feature engineering.
[0013] Another objective of the present disclosure is to recognize complex, overlapping and improving performance in dynamic settings.
[0014] Another objective of the present disclosure is to handle noisy, missing, and inconsistent sensor data effectively, maintaining high recognition accuracy under real-world conditions.
[0015] Another objective of the present disclosure is to provide actionable outputs, such as alerts, notifications, based on recognized activities.
[0016] Another objective of the present disclosure is to enable scalable deployment in smart homes, healthcare monitoring, security systems, and fitness applications, supporting both single-user and multi-user environments.
[0017] Another objective of the present disclosure is to enhance the generalization of activity recognition models, enabling robust performance across various environments, sensor types, and user populations.
[0018] Yet another objective of the present disclosure is to provide a commercial solution that balances high recognition accuracy, energy efficiency, privacy, and ease of deployment, offering users a reliable alternative to existing Human Recognition systems.
[0019] In light of the above, in one aspect of the present disclosure, a deep learning-based human activity recognition system is disclosed herein. The system comprises a user device integrated a plurality of cameras and configured to capture high-resolution images and videos for real-time monitoring, analysing movements and behaviours from a user through a user interface. The system includes a plurality of sensors positioned within the user device and configured to sense motion, location, orientation, and environmental conditions of a user. The system also includes a communication network configured to establish a communication link for data transmission within the system. The system further includes a processing unit connected to the user device, the plurality of cameras, and the plurality of sensors via the communication network, the processing unit configured to process real-time data for human activity recognition, wherein the processing unit further comprises a data input module configured to receive real-time data from the plurality of cameras, and the plurality of sensors positioned within the user device, a pre-processing module configured to clean, normalize and reduce noise in the received data to enhance the data quality, a feature extraction module configured to extract relevant features from the pre-processed data including motion patterns, gestures, postures, and other relevant feature parameters, a human activity detection module configured to analyse the extracted features and determine the activities performed by the user, a human activity classification module configured to recognize and classify the detected human activity into predefined categories including presence of motion, gestures, change in postures, object interactions, cyclic actions and multi-user activities, an output module configured to transmit the extracted features, detected human activity and classified human activity to the user device.
[0020] In one embodiment, the cloud database configured to store and managed data associated with human activity recognition, thereby enabling remote access and improved classification accuracy.
[0021] In one embodiment, the plurality of sensors comprises an accelerometers and a gyroscopes configured to sense motion, orientation, and environmental conditions of a user for real-time human activity recognition.
[0022] In one embodiment, the pre-processing module incorporates a dynamic sensor reliability weighting engine that evaluates the real-time performance of the plurality of sensors and the plurality of cameras configured to scale the corresponding sensor data according to its reliability, so that the weighted sensor data is provided as input to a deep learning model, thereby improving the accuracy and robustness of activity recognition and classification.
[0023] In one embodiment, the human activity detection module configured to determine and track the activities of multiple users within the same environment, thereby enabling accurate recognition of overlapping human activities including but not limited to simultaneous gestures, interacting tasks, cyclic motions, and concurrent movement patterns.
[0024] In one embodiment, the human activity classification module applies a trained deep learning model selected from artificial neural network, convolutional neural network, recurrent neural network, and a hybrid architecture to classify user activities in real-time.
[0025] In one embodiment, the processing unit further comprises an alert generation module configured to generate a real-time alerts upon detection of unsafe activities including but not limited to sudden falls, prolonged inactivity, abnormal movement patterns and unauthorised access into restricted areas.
[0026] In one embodiment, the processing unit further comprises a training and testing module configured to split the data into training and testing data sets to train the system, evaluates its performance, improves the system stability and interpretability.
[0027] In light of the above, in one aspect of recent disclosure a method of making a deep learning-based human activity recognition system is disclosed herein. The method comprises capturing high-resolution images and videos for real-time monitoring, analysing movements and behaviours from a user through a user interface via the user device. The method also includes sensing motion, location, orientation, and environmental conditions of a user via the plurality of sensors. The method further includes establishing a communication link for data transmission within the system via the communication network. The method also includes processing real-time data for human activity recognition via the processing unit receiving real-time data from the plurality of cameras, and the plurality of sensors positioned within the user device via the data input module. The method includes cleaning, normalize and reduce noise in the received data to enhance the data quality via the pre-processing module. The method also includes extracting relevant features from the pre-processed data including motion patterns, gestures, postures, and other relevant feature parameters via the feature extraction module. The method includes analysing the extracted features and determine the activities performed by the user via the human activity detection module. The method further includes measure the duration of each detected activity and provide this information to the human activity classification module to improve the accuracy of classifying activities exhibiting time-dependent patterns, such as repeated motions, prolonged inactivity, and sequential tasks. The method also includes recognizing and classify the detected human activity into predefined categories including presence of motion, gestures, change in postures, object interactions, cyclic actions and multi-user activities via the human activity classification module. The method further includes generating real-time alerts upon detection of unsafe activities including but not limited to sudden falls, prolonged inactivity, abnormal movement patterns and unauthorised access into restricted areas via the alert generation module. The method includes splitting the data into training and testing data sets to train the system, evaluates its performance, improves the system stability and interpretability via the training and testing module. Moreover, the method also includes transmitting the extracted features, detected human activity and classified human activity to the user device via the output module. Furthermore, the method also includes displaying extracted feature, detected human activity, classified human activity on the user interface of the user device.
[0028] These and other advantages will be apparent from the present application of the embodiments described herein.
[0029] The preceding is a simplified summary to provide an understanding of some embodiments of the present invention. This summary is neither an extensive nor exhaustive overview of the present invention and its various embodiments. The summary presents selected concepts of the embodiments of the present invention in a simplified form as an introduction to the more detailed description presented below. As will be appreciated, other embodiments of the present invention are possible utilizing, alone or in combination, one or more of the features set forth above or described in detail below.
[0030] These elements, together with the other aspects of the present disclosure and various features are pointed out with particularity in the claims annexed hereto and form a part of the present disclosure. For a better understanding of the present disclosure, its operating advantages, and the specified object attained by its uses, reference should be made to the accompanying drawings and descriptive matter in which there are illustrated exemplary embodiments of the present disclosure.
BRIEF DESCRIPTION OF THE DRAWINGS
[0031] To describe the technical solutions in the embodiments of the present disclosure or in the prior art more clearly, the following briefly describes the accompanying drawings required for describing the embodiments or the prior art. Apparently, the accompanying drawings in the following description merely show some embodiments of the present disclosure, and a person of ordinary skill in the art can derive other implementations from these accompanying drawings without creative efforts. All of the embodiments or the implementations shall fall within the protection scope of the present disclosure.
[0032] The advantages and features of the present disclosure will become better understood with reference to the following detailed description taken in conjunction with the accompanying drawing, in which:
[0033] FIG. 1 illustrates a block diagram of a deep learning-based human activity recognition system, in accordance with an embodiment of the present disclosure;
[0034] FIG. 2 illustrates a flowchart of a method, outlining the sequential steps for the deep learning-based human activity recognition system, in accordance with an embodiment of the present disclosure;
[0035] Like reference, numerals refer to like parts throughout the description of several views of the drawing.
[0036] The deep learning-based human activity recognition system is illustrated in the accompanying drawings, which like reference letters indicate corresponding parts in the various figures. It should be noted that the accompanying figure is intended to present illustrations of exemplary embodiments of the present disclosure. This figure is not intended to limit the scope of the present disclosure. It should also be noted that the accompanying figure is not necessarily drawn to scale.
DETAILED DESCRIPTION OF THE DISCLOSURE
[0037] The following is a detailed description of embodiments of the disclosure depicted in the accompanying drawings. The embodiments are in such detail as to communicate the disclosure. However, the amount of detail offered is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the scope of the present disclosure.
[0038] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present disclosure. It may be apparent to one skilled in the art that embodiments of the present disclosure may be practiced without some of these specific details.
[0039] Various terms as used herein are shown below. To the extent a term is used, it should be given the broadest definition persons in the pertinent art have given that term as reflected in printed publications and issued patents at the time of filing.
[0040] The terms “a” and “an” herein do not denote a limitation of quantity, but rather denote the presence of at least one of the referenced items.
[0041] The terms “having”, “comprising”, “including”, and variations thereof signify the presence of a component.
[0042] Referring now to FIG. 1 to FIG. 2 to describe various exemplary embodiments of the present disclosure. FIG. 1 illustrates a deep learning-based human activity recognition system, in accordance with an embodiment of the present disclosure.
[0043] The system 100 may include a user device 102. The system 100 may include a plurality of cameras 104. The system 100 may include a user interface 106. The system 100 may include a plurality of sensors 108. The system 100 may include an accelerometers 110. The system 100 may include a gyroscopes 112. The system 100 may include a communication network 114. The system 100 may include a processing unit 116, further comprises a data input module 118, a pre-processing module 120, a dynamic sensor reliability weighting engine 122, a feature extraction module 124, a human activity detection module 126, a human activity classification module 128, an alert generation module 130, a training and testing module 132, an output module 134. The system 100 may include a cloud database 136.
[0044] The user device 102 integrated a plurality of cameras 104 and configured to capture high-resolution images and videos for real-time monitoring, analysing movements and behaviours from a user through a user interface 106. The user device 102 may be implemented as a smartphone, a tablet, a dedicated edge artificial intelligence processing unit, a wearable device, and an integrated smart surveillance unit.
[0045] In one embodiment of the present invention, the user device 102 further includes dedicated edge computing device designed for on-device inference, thereby reducing latency and enhancing privacy by minimizing cloud dependency.
[0046] In one embodiment of the present invention, the plurality of cameras 104 further includes visible spectrum cameras, infrared cameras, depth-sensing cameras, and wide-angle cameras positioned strategically to maximize field-of-view coverage. The plurality of cameras 104 operate at adjustable frame rates and resolutions to optimize between image clarity and processing efficiency based on the system 100 requirements.
[0047] In one embodiment of the present invention, the user interface 106 configured to display live video feeds, detected activity labels, behavioural alerts, and the system 100 notifications. The user interface 106 may include a touchscreen display, mobile application interface, web-based dashboard, and voice-enabled control module. The user interface 106 further allows configuration of monitoring parameters, privacy settings, camera calibration, and alert thresholds.
[0048] The plurality of sensors 108 positioned within the user device 102 and configured to sense motion, location, orientation, and environmental conditions of a user. The plurality of sensors 108 further include location-determining modules such as Global Positioning System receivers, indoor positioning sensors, Bluetooth-based proximity sensors, and ultra-wideband positioning components configured to determine the spatial position of the user within indoor and outdoor environments. Such location information enables contextual awareness and zone-based activity recognition.
[0049] In one embodiment of the present invention, the plurality of sensors 108 comprises an accelerometers 110 and a gyroscopes 112 configured to sense motion, orientation, and environmental conditions of a user for real-time human activity recognition.
[0050] In one embodiment of the present invention, the plurality of sensors 108 further includes accelerometers 110 configured to measure linear acceleration along multiple axes and gyroscopes 112 configured to detect angular velocity and rotational movement.
[0051] In one embodiment of the present invention, the plurality of sensors 108 further includes magnetometers configured to determine directional heading relative to the Earth’s magnetic field. These motion-related sensors enable detection of user activities such as walking, sitting, standing, falling, bending, and abrupt movements.
[0052] In one embodiment of the present invention, the plurality of sensors 108 further includes environmental sensors including temperature sensors, humidity sensors, ambient light sensors, barometric pressure sensors, and air-quality sensors. These environmental sensors provide contextual data that enhances activity interpretation, such as distinguishing between indoor and outdoor activities, detecting unusual environmental conditions, and identifying emergency situations.
[0053] The communication network 114 configured to establish a communication link for data transmission within the system 100 and provides connectivity between components of the system 100, enabling data exchange, control signalling, and remote communication.
[0054] In one embodiment of the present invention, the communication network 114 may include but not limited to wired and wireless networks.
[0055] In one embodiment of the present invention, the communication network 114 includes wireless networks including but not limited to Bluetooth and wireless fidelity.
[0056] The processing unit 116 connected to the user device 102, the plurality of cameras 104, and the plurality of sensors 108 via the communication network 114, the processing unit 116 configured to process real-time data for human activity recognition. The processing unit 116 is implemented as an edge computing module configured to perform on-device inference to reduce latency and enhance privacy.
[0057] In one embodiment of the present invention, the processing unit 116 may support adaptive learning mechanisms, enabling incremental personalization based on newly acquired user data, while maintaining data security through encryption and secure communication protocols. Accordingly, the processing unit 116 enables efficient, real-time, and intelligent human activity recognition within the Internet of Things-enabled smart environment.
[0058] The data input module 118 configured to receive real-time data from the plurality of cameras 104, and the plurality of sensors 108 positioned within the user device 102. The data input module 118 functions as an interface layer responsible for acquiring, buffering, synchronizing, and routing multimodal data also support multiple input formats including video frames, image sequences, time-series sensor signals, and environmental readings.
[0059] The pre-processing module 120 configured to clean, normalize and reduce noise in the received data to enhance the data quality. The pre-processing module 120 performs signal conditioning operations including filtering, smoothing, interpolation of missing values, outlier detection, frame stabilization for video streams, and removal of corrupted and redundant data samples.
[0060] In one embodiment of the present invention, the pre-processing module 120 incorporates a dynamic sensor reliability weighting engine 122 that evaluates the real-time performance of the plurality of sensors 108 and the plurality of cameras 104 and configured to scale the corresponding sensor data according to its reliability, so that the weighted sensor data is provided as input to a deep learning model, thereby improving the accuracy and robustness of activity recognition and classification.
[0061] In one embodiment of the present invention, the pre-processing module 120 further includes dynamic sensor reliability weighting engine 122 continuously monitors performance indicators such as signal-to-noise ratio, packet loss rate, sampling consistency, battery level, occlusion detection in cameras, lighting conditions, motion blur, and environmental interference. Based on these indicators, the engine assigns a dynamic reliability score to each sensor and camera stream.
[0062] In one embodiment of the present invention, the pre-processing module 120 further includes dynamic sensor reliability weighting engine 122 may employ statistical analysis, rule-based evaluation, and machine learning-based confidence estimation models to compute the reliability scores. The weighting process may occur in real-time and adapt continuously to changing environmental conditions and the system 100 performance variations.
[0063] The feature extraction module 124 configured to extract relevant features from the pre-processed data including motion patterns, gestures, postures, and other relevant feature parameters. The feature extraction module 124 computes statistical, temporal, and frequency-domain features including mean, variance, standard deviation, root mean square, signal magnitude area, zero-crossing rate, entropy, correlation coefficients between axes, and energy distribution across frequency bands.
[0064] In one embodiment of the present invention, the feature extraction module 124 further includes skeletal pose estimation algorithms are applied to detect and track key body joints, generating joint coordinate maps and limb angle measurements that represent posture and gesture information. Optical flow analysis may be implemented to determine motion direction and velocity between successive frames, enabling recognition of dynamic gestures and locomotion activities.
[0065] In one embodiment of the present invention, the feature extraction module 124 further includes posture-related features such as body inclination angle, sitting versus standing alignment, fall trajectory characteristics, limb extension patterns, and repetitive motion signatures. Contextual features incorporating environmental sensor inputs such as ambient light level, temperature variation, and location zone identification may also be integrated to enhance semantic interpretation of activities.
[0066] The human activity detection module 126 configured to analyse the extracted features and determine the activities performed by the user. The human activity detection module 126 performs both activity classification and intelligent activity detection, enabling accurate, real-time recognition of human behaviour in Internet of Things-enabled smart environments.
[0067] In one embodiment of the present invention, the human activity detection module 126 configured to determine and track the activities of multiple users within the same environment, thereby enabling accurate recognition of overlapping human activities including but not limited to simultaneous gestures, interacting tasks, cyclic motions, and concurrent movement patterns.
[0068] In one embodiment of the present invention, the human activity detection module 126 further includes higher-level detection functionalities such as temporal smoothing to reduce rapid fluctuations in activity labels, sequence modelling to recognize overlapping and complex activities, multi-user differentiation for distinguishing simultaneous actions of multiple individuals, and anomaly detection to identify irregular activity patterns, such as falls, prolonged inactivity, or sudden deviations from learned behavioural patterns.
[0069] The human activity classification module 128 configured to recognize and classify the detected human activity into predefined categories including presence of motion, gestures, change in postures, object interactions, cyclic actions and multi-user activities. The human activity classification module 128 operates as the core classification engine, mapping extracted features and reliability-weighted data to specific activity labels using trained deep learning models.
[0070] In one embodiment of the present invention, the human activity classification module 128 applies a trained deep learning model selected from artificial neural network, convolutional neural network, recurrent neural network, and a hybrid architecture to classify user activities in real-time.
[0071] In one embodiment of the present invention, the human activity classification module 128 further employs an Artificial Neural Network as the primary classification model for predicting user activities in Internet of Things-enabled smart homes. Experimental results on the Activity Recognition Ambient Sensors dataset demonstrate that the Artificial Neural Network model outperforms alternative deep learning architectures such as Convolutional Neural Networks and Recurrent Neural Networks.
[0072] In one embodiment of the present invention, the human activity classification module 128 incorporates confidence scoring and probabilistic output, assigning each activity label a confidence score that represents the model’s certainty. Temporal analysis and sequence modelling may be applied to smooth predictions over time, reducing transient misclassifications and improving reliability.
[0073] In one embodiment of the present invention, the human activity classification module 128 improve model performance, the input sensor and camera data are subjected to feature scaling prior to training. To attain the best level of accuracy using the selected deep learning models, the datasets were segmented into zeros (0) and ones (1) by using MinMaxScaler,
The values that exist prior to scaling are Xi, Xmin, and Xmax. Grouping the created models were assessed for accuracy, recall, precision, and F1-measure. Evaluation matrices quantify how effectively the model projected outcome variables using newly collected, unlabelled data.
[0074] In one embodiment of the present invention, the human activity classification module 128 further includes accuracy is calculated as the ratio of true positives to the total number of predictions, accounting for false positives and false negatives.

[0075] In one embodiment of the present invention, the human activity classification module 128 further includes precision represents the proportion of correctly predicted positive samples out of all predicted positives.

[0076] In one embodiment of the present invention, the human activity classification module 128 further includes recall is defined as the proportion of correctly predicted positive samples out of all actual positives.

[0077] In one embodiment of the present invention, the human activity classification module 128 further includes F-1 measure Average accuracy (F-1) recall. Unit of measurement. The inverse of True positives are True negatives, False positives, and False negatives.

[0078] The alert generation module 130 configured to generate a real-time alerts upon detection of unsafe activities including but not limited to sudden falls, prolonged inactivity, abnormal movement patterns and unauthorised access into restricted areas.
[0079] In one embodiment of the present invention, the alert generation module 130 may implement threshold-based, rule-based, and probabilistic anomaly detection algorithms to determine the observed activity warrants an immediate alert.
[0080] In one embodiment of the present invention, the alert generation module 130 may include smartphone notifications, emails, Short Message Services messages, in-home smart speakers and visual displays.
[0081] The training and testing module 132 configured to split the data into training and testing data sets to train the system 100, evaluates its performance, improves the system 100 stability and interpretability. The training and testing module 132 supports incremental learning, enabling periodic retraining with newly acquired user data while maintaining model interpretability and preventing catastrophic forgetting.
[0082] The output module 134 configured to transmit the extracted features, detected human activity and classified human activity to the user device 102. The output module 134 formats the information for real-time display, storage, and further processing. This may include streaming activity status to a user interface 106 on the user device 102, updating dashboards, and sending data to cloud-based systems for archival and remote monitoring.
[0083] FIG. 2 illustrates a flowchart of a method, outlining the sequential steps for the deep learning-based human activity recognition system 100, in accordance with an embodiment of the present disclosure;
[0084] At step 202, the high-resolution images and videos is captured for real-time monitoring, analysing movements and behaviours from a user through a user interface 106 via the user device 102.
[0085] At step 204, the motion, location, orientation, and environmental conditions is sensed of a user via the plurality of sensors 108.
[0086] At step 206, the communication link is established for data transmission within the system 100 via the communication network 114.
[0087] At step 208, the real-time data is processed for human activity recognition via the processing unit 116, comprising several modules.
[0088] At step 210, the real-time data is received from the plurality of cameras 104, and the plurality of sensors 106 positioned within the user device 102 via the data input module 118.
[0089] At step 212, the received data is cleaned, normalize and reduce noise to enhance the data quality via the pre-processing module 120.
[0090] At step 214, the relevant features is extracted from the pre-processed data including motion patterns, gestures, postures, and other relevant feature parameters via the feature extraction module 124.
[0091] At step 216, the extracted features is analysed and determine the activities performed by the user via the human activity detection module 126.
[0092] In one embodiment of the present invention, the human activity detection module 126 configured to measure the duration of each detected activity and provide this information to the human activity classification module 128 to improve the accuracy of classifying activities exhibiting time-dependent patterns, such as repeated motions, prolonged inactivity, and sequential tasks.
[0093] At step 218, the detected human activity is recognised and classified into predefined categories including presence of motion, gestures, change in postures, object interactions, cyclic actions and multi-user activities via the human activity classification module 128.
[0094] At step 220,the real-time alerts is generated upon detection of unsafe activities including but not limited to sudden falls, prolonged inactivity, abnormal movement patterns and unauthorised access into restricted areas via the alert generation module 130.
[0095] At step 222, the data is splitted into training and testing data sets to train the system 100, evaluates its performance, improves the system 100 stability and interpretability via the training and testing module 132.
[0096] At step 224, the extracted features, detected human activity and classified human activity is transmitted to the user device 102 via the output module 134.
[0097] At step 226, the extracted feature, detected human activity, classified human activity is displayed on the user interface 106 of the user device 102.
[0098] In the best mode of present invention, the system 100 operates to provide real-time, accurate, and robust human activity recognition in Internet of Things-enabled smart environments. The system 100 captures high-resolution images and video streams of a user through the plurality of cameras 104 integrated into the user device 102, while simultaneously collecting motion, orientation, location, and environmental data from the plurality of sensors 108, including accelerometers 110 and gyroscopes 112. The collected multimodal data is transmitted via the communication network 114 to the processing unit 116, wherein the data input module 118 receives and forwards it to the pre-processing module 120 for cleaning, normalization, noise reduction, and dynamic feature scaling. The pre-processing module 120 incorporates a dynamic sensor reliability weighting engine 122, which continuously evaluates the real-time performance of each sensor and camera, assigning weights to prioritize high-quality data, thereby improving the accuracy and robustness of subsequent processing. The feature extraction module 124 then analyses the pre-processed and weighted data to generate meaningful features representing motion patterns, gestures, postures, object interactions, cyclic actions, and contextual environmental indicators. These features are processed by the human activity detection module 126, which identifies ongoing activities, tracks multi-user and overlapping activities, and measures activity duration to capture time-dependent patterns. The human activity classification module 128 applies a trained deep learning model, such as an Artificial Neural Network, Convolutional Neural Network, Recurrent Neural Network, Long Short-Term Memory Network, or hybrid architecture, to classify the detected activities into predefined categories, including presence of motion, gestures, changes in posture, object interactions, cyclic actions, and multi-user activities, while incorporating confidence scores and temporal sequences to improve reliability and reduce misclassification. Upon detecting unsafe activities, such as sudden falls, prolonged inactivity, abnormal movement patterns, and unauthorized access to restricted areas, the alert generation module 130 generates real-time notifications to the user device 102 and remote monitoring systems, and the output module 134 transmits the extracted features, detected activity, and classified activity to the user interface 106 for visualization, monitoring, and feedback. The training and testing module 132 ensures optimal system 100 performance by splitting datasets into training and testing subsets, training the deep learning models, evaluating performance using accuracy, precision, recall, and F1-measure metrics, and iteratively refining model parameters, while supporting incremental learning from new user data to maintain adaptability and personalization. In this configuration, the system 100 provides an end-to-end solution for real-time, reliable, and context-aware human activity recognition, enhancing safety, security, and automation in smart home and Internet of Things-enabled environments.
[0099] While the invention has been described in connection with what is presently considered to be the most practical and various embodiments, it will be understood that the invention is not to be limited to the disclosed embodiments, but on the contrary, is intended to cover various modifications and equivalent arrangements included within the scope of the appended claims.
[0100] A person of ordinary skill in the art may be aware that, in combination with the examples described in the embodiments disclosed in this specification, units and algorithm steps may be implemented by electronic hardware, computer software, or a combination thereof.
[0101] The foregoing descriptions of specific embodiments of the present disclosure have been presented for purposes of illustration and description. They are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed, and many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described to best explain the principles of the present disclosure and its practical application, and to thereby enable others skilled in the art to best utilize the present disclosure and various embodiments with various modifications as are suited to the particular use contemplated. It is understood that various omissions and substitutions of equivalents are contemplated as circumstances may suggest or render expedient, but such omissions and substitutions are intended to cover the application or implementation without departing from the scope of the present disclosure.
[0102] Disjunctive language such as the phrase “at least one of X, Y, Z,” unless specifically stated otherwise, is otherwise understood with the context as used in general to present that an item, term, etc., may be either X, Y, or Z, or any combination thereof (e.g., X, Y, and/or Z). Thus, such disjunctive language is not generally intended to, and should not, imply that certain embodiments require at least one of X, at least one of Y, or at least one of Z to each be present.
[0103] In a case that no conflict occurs, the embodiments in the present disclosure and the features in the embodiments may be mutually combined. The foregoing descriptions are merely specific implementations of the present disclosure but are not intended to limit the protection scope of the present disclosure. Any variation or replacement readily figured out by a person skilled in the art within the technical scope disclosed in the present disclosure shall fall within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be subject to the protection scope of the claims.
, Claims:I/We Claim:
1. A deep learning-based human activity recognition system (100), the system (100) comprising:
a user device (102) integrated a plurality of cameras (104) and configured to capture high-resolution images and videos for real-time monitoring, analysing movements and behaviours from a user through a user interface (106);
a plurality of sensors (108) positioned within the user device (102) and configured to sense motion, location, orientation, and environmental conditions of a user;
a communication network (114) configured to establish a communication link for data transmission within the system (100);
a processing unit (116) connected to the user device (102), the plurality of cameras (104), and the plurality of sensors (108) via the communication network (114), the processing unit (116) configured to process real-time data for human activity recognition, wherein the processing unit (116) further comprises:
a data input module (118) configured to receive real-time data from the plurality of cameras (104), and the plurality of sensors (108) positioned within the user device (102);
a pre-processing module (120) configured to clean, normalize and reduce noise in the received data to enhance the data quality;
a feature extraction module (124) configured to extract relevant features from the pre-processed data including motion patterns, gestures, postures, and other relevant feature parameters;
a human activity detection module (126) configured to analyse the extracted features and determine the activities performed by the user;
a human activity classification module (128) configured to recognize and classify the detected human activity into predefined categories including presence of motion, gestures, change in postures, object interactions, cyclic actions and multi-user activities;
an output module (134) configured to transmit the extracted features, detected human activity and classified human activity to the user device (102).
2. The system (100) as claimed in claim 1, wherein the system (100) further comprises a cloud database (136) configured to store and managed data associated with human activity recognition, thereby enabling remote access and improved classification accuracy.
3. The system (100) as claimed in claim 1, wherein the plurality of sensors (108) comprises an accelerometers (110) and a gyroscopes (112) configured to sense motion, orientation, and environmental conditions of a user for real-time human activity recognition.
4. The system (100) as claimed in claim 1, wherein the pre-processing module (120) incorporates a dynamic sensor reliability weighting engine (122) that evaluates the real-time performance of the plurality of sensors (108) and the plurality of cameras (104) configured to scale the corresponding sensor data according to its reliability, so that the weighted sensor data is provided as input to a deep learning model, thereby improving the accuracy and robustness of activity recognition and classification.
5. The system (100) as claimed in claim 1, wherein the human activity detection module (126) configured to determine and track the activities of multiple users within the same environment, thereby enabling accurate recognition of overlapping human activities including but not limited to simultaneous gestures, interacting tasks, cyclic motions, and concurrent movement patterns.
6. The system (100) as claimed in claim 1, wherein the human activity classification module (128) applies a trained deep learning model selected from artificial neural network, convolutional neural network, recurrent neural network, and a hybrid architecture to classify user activities in real-time.
7. The system (100) as claimed in claim 1, wherein the processing unit (116) further comprises an alert generation module (130) configured to generate a real-time alerts upon detection of unsafe activities including but not limited to sudden falls, prolonged inactivity, abnormal movement patterns and unauthorised access into restricted areas.
8. The system (100) as claimed in claim 1, wherein the processing unit (116) further comprises a training and testing module (132) configured to split the data into training and testing data sets to train the system (100), evaluates its performance, improves the system (100) stability and interpretability.
9. A method (200) for deep learning-based human activity recognition system (100), the method (200) comprising:
capturing high-resolution images and videos for real-time monitoring, analysing movements and behaviours from a user through a user interface (106) via the user device (102);
sensing motion, location, orientation, and environmental conditions of a user via the plurality of sensors (108);
establishing a communication link for data transmission within the system (100) via the communication network (114);
processing real-time data for human activity recognition via the processing unit (116), comprising several modules;
receiving real-time data from the plurality of cameras (104), and the plurality of sensors (106) positioned within the user device (102) via the data input module (118);
cleaning, normalize and reduce noise in the received data to enhance the data quality via the pre-processing module (120);
extracting relevant features from the pre-processed data including motion patterns, gestures, postures, and other relevant feature parameters via the feature extraction module (124);
analysing the extracted features and determine the activities performed by the user via the human activity detection module (126);
recognizing and classify the detected human activity into predefined categories including presence of motion, gestures, change in postures, object interactions, cyclic actions and multi-user activities via the human activity classification module (128);
generating real-time alerts upon detection of unsafe activities including but not limited to sudden falls, prolonged inactivity, abnormal movement patterns and unauthorised access into restricted areas via the alert generation module (130);
splitting the data into training and testing data sets to train the system (100), evaluates its performance, improves the system (100) stability and interpretability via the training and testing module (132);
transmitting the extracted features, detected human activity and classified human activity to the user device (102) via the output module (134);
displaying extracted feature, detected human activity, classified human activity on the user interface (106) of the user device (102).
10. The system (100) as claimed in claim 1, wherein the Human Activity Detection Module (126) configured to measure the duration of each detected activity and provide this information to the human activity classification module (128) to improve the accuracy of classifying activities exhibiting time-dependent patterns, such as repeated motions, prolonged inactivity, and sequential tasks.