VIDEO ANALYTICS-BASED BELT CONVEYOR OPERATION
FIELD OF THE INVENTION
The present invention is related to a video analytics and deep learning-based systems to that enable remote monitoring of a belt conveyor operation, more specifically, a video analytics system that uses Artificial Intelligence and Machine Learning (AIML) tools to improve the surveillance and quick response to any deviations with respect to normal running conditions.
BACKGROUND OF THE INVENTION
Background description includes information that may be useful in understanding the present invention. It is not an admission that any of the information provided herein is prior art or relevant to the presently claimed invention, or that any publication specifically or implicitly referenced is prior art.
In prior art technologies, belt conveyor operation is being monitored by a Belt conveyor local area operator, who ensures that the Belt conveyor operates with-in normal operating conditions and any deviation such as, idle running, coal/ material on belt in stop condition, belt running misaligned, vicinity of human beings, etc., are to be reported immediately to control room engineer. However, it’s almost an impossible task for human operators to effectively keep a check on the machinery 24/7. Moreover, the operator may fail to observe some minor or crucial aspects regarding the working that may lead to failure of the machinery. It’s also not possible for a human operator to forecast efficient working of the machinery in the long run, and therefore, may find it difficult for him/her to determine the condition of the belt conveyor operation and to detect a potential failure in advance. This results in high maintenance, increased downtime of conveyors, and a considerable decrease in efficiency of the conveyors and production in the industry. Therefore, there is a need for change in these above-mentioned aspects with regard to detection of belt damages, human safety, wear and tear, insufficient cleaning, ripping off the conveyor belt, and detection of foreign objects, which becomes essential to avoid any damages to the conveyor belt that may result in costly shutdowns.
In view of the above, there is a long felt but unforeseen need for a method or a system that performs the activities of an operator who operates a belt conveyor system, in a more efficient and substantial manner. In other words, there is a need for a method and system that

visually monitors, for example, a video analytics and deep learning-based application that enables remote monitoring of belt conveyor operation by using Artificial Intelligence and Machine Learning (AIML) tools to improve the surveillance and quick response to any deviations with respect to normal running conditions.
SUMMARY OF THE INVENTION
It is intended that all such features, and advantages be included within this description, be within the scope of the present invention. The following summary is provided to facilitate an understanding of some of the innovative features unique to the disclosed embodiment and is not intended to be a full description. A full appreciation of the various aspects of the embodiments disclosed herein can be gained by taking the entire specification, drawings, and abstract as a whole.
The primary objective of the present invention is to develop a video analytics-based model that detects process anomalies of coal conveying operations, for example, idle running of conveyor, loaded stop condition, belt misalignment and human detection in vicinity of belt conveyor. A video analytics-based belt conveyor system disclosed herein addresses the above-mentioned need for a method and system that visually monitors, for example, a video analytics and deep learning-based application that enables remote monitoring of belt conveyor operation by using Artificial Intelligence and Machine Learning (AIML) tools to improve the surveillance and quick response to any deviations with respect to normal running conditions.
The video analytics-based belt conveyor system disclosed herein comprises a camera, a computing device that comprises a memory storage and a hardware processor, and a model. The camera is placed in a proximity of a conveyor belt to continuously transmit real time feed of video data to the computing device that comprises the memory storage and the hardware processor. The memory storage comprises stored instructions that when executed cause the hardware processor to perform a plurality of operations. The hardware processor converts the video data into multiple frames. The model that is generated based on a machine learning method and stored in the memory storage of the computing device compares the frames and checks for anomalies in working of the conveyor belt based on characteristics of each frame. In response to an identified anomaly, information on the identified anomaly is detected by the

hardware processor of the computing device and the hardware processor generates an alarm based on the identified anomaly to alert a user regarding the working of the conveyor belt.
In an embodiment, if there is no anomaly detected during the anomaly check, then the hardware processor continuously evaluates real time feed of subsequent video data. In an embodiment, multiple cameras are positioned in the proximity of multiple conveyor belts to continuously transmit the real time feed of the video data to the computing device, where the hardware processor retrieves the frames from the video data from each camera and analyses each frame of the video data separately to identify the anomalies and determines number of the alerts to be generated based on the anomalies. In an embodiment, the hardware processor is configured to: collect real time feed of the video data from the camera; extract the frames from the video data, which is saved into a folder; label the frames from the video data into different classes using a labelling tool; pre-process the labelled data, wherein the pre-processing comprises cropping the frames, converting the frames into grey scale, and scaling the frames; train the model using the pre-processed data and save the trained data in the memory storage; hyper parameter tune the trained data to extract the best parameters; test the model to facilitate accuracy of the model in a real-world data; deploy the model on a server using an interface; predict the frames using the interface in real time; and generate an alert.
In an embodiment, the machine learning method is a Convolutional Neural Network (CNN) based deep learning algorithm, which comprises steps to: receive, via an input layer, the frames from the video data; normalize each frame so that video data is readable by the CNN algorithm; define the model using the normalized frames; apply filters using a first convolution layer to create a feature map; perform, via a first pooling layer, a reduction in dimensions of the normalized frames; apply filters using a second convolution layer to create another feature map; perform, via a second pooling layer, another reduction in dimensions of the normalized frames; apply filters using a third convolution layer to create another feature map; and choose a maximum output, via SoftMax, in the form of the alarm that needs to be generated based on the frames.
In short, the present invention is related to an intelligent video analytic system that consists of different architecture components which includes data capturing-recording, analysing, live feed optimizing, image modelling, model training, video analytics algorithm with open-source libraries/ open cv-keras, tensor flow and process output for taking intelligent decision

making. This system continuously analyses the video feed and provide intelligent output to control room engineer for taking operational decision.
USE CASES:
1. Idle running of conveyor- The video analytics-based belt conveyor system is used to overcome energy and productivity loss when the conveyor is running but not conveying coal.
2. Loaded Stop of conveyor - The video analytics-based belt conveyor system is used to prevent a fire hazard due to self- combustion property (smouldering) when the conveyor is loaded with coal but not running.
3. Chute Choke up - The video analytics-based belt conveyor system is used to further prevent interruption in coal feeding operation and production loss when the discharge chute of one conveyor and receiving chute of subsequent conveyor is jammed with coal due to coal not being conveyed by next conveyor while first one keep discharging. Here, the video analytics-based belt conveyor system also helps in reducing additional man efforts required to clear choke up.
The video analytics-based belt conveyor system is basically used to detect such above mentioned abnormalities or anomalies using video analytics-based machine learning tools (AI). Idle running is detected by a matrix which is a combination of two different scenario on the same frame/video feed. First is whether the conveyor is “Running or Not Running” and the second is whether the conveyor “ Having Coal or No Coal”. Detecting whether a conveyor is running or not running is identified using the frame (state of both coal or no coal) that is retrieved by the light linked with the conveyor motion.
Since the colour of conveyor belt and coal is similar, and therefore, the evaluation of condition “ coal or no coal” on conveyor belt is derived by deep learning concept in both cases “conveyor running or not running”. Conveyor “running or not running” and “coal or no coal” on the conveyor belt are based on four combination cases, out of which two are normal conditions and two are abnormality.
A. Conveyor running and with coal - normal operation condition.
B. Conveyor not running and with no coal - normal stop condition.
C. Conveyor running and no coal - idle running condition – abnormality.
D. Conveyor not running and with coal- loaded stop condition – abnormality.

E. Chute choke is detected by combination of two conveyor’s operating condition i.e. First conveyor “running with coal” and next conveyor “no coal and running”.
4. Conveyor belt mis-alignment- it means belt is not running in the centre. This causes rubbing of belt to structure, belt edge damage, energy loss and interruption in feeding on tripping on PLC based control. This abnormality is detected using the image analytics that’s formed using the video analytics-based belt conveyor system, which uses image processing with which an edge of the conveyor belt is detected and predicted whether the edges are aligned or mis-aligned.
5. Human detection in vicinity of conveyor- The video analytics-based belt conveyor system detects humans in the vicinity of the running conveyor, which presents a high risk of entrapment that may lead to severe injury. This abnormality is detected use deep learning technique (CNN).
BRIEF DESCRIPTION OF DRAWINGS
The invention can be better understood with reference to the following drawings. The components in the drawings are not necessarily to scale, emphasis instead being placed upon clearly illustrating the principles of the present invention. Moreover, in the drawings, like reference numerals designate corresponding parts throughout the several views.
Figure 1A exemplarily illustrates a basic flow diagram of the operation of the video analytics-based belt conveyor system, as an example embodiment of the present invention.
Figure 1B exemplarily illustrates a phase 1 architecture of the operation of the video analytics-based belt conveyor system.
Figure 1C exemplarily illustrates an overall architecture for multiple cameras in the operation of the video analytics-based belt conveyor system.
Figure 1D exemplarily illustrates a camera configuration table for the operation of the video analytics-based belt conveyor system.

Figure 1E exemplarily illustrates a camera transaction table for the operation of the video analytics-based belt conveyor system.
Figure 1F exemplarily illustrates a model processing table for the operation of the video analytics-based belt conveyor system.
Figure 2 exemplarily illustrates an application workflow of the operation of the video analytics-based belt conveyor system.
Figure 3 exemplarily illustrates a Convolutional Neural Network (CNN) based deep learning algorithm of the video analytics-based belt conveyor operation.
Figure 4 exemplarily illustrates an object detection model of the video analytics-based belt conveyor system, as disclosed in Figures 1-5.
Figures 5A-5E exemplarily illustrate an Object Detection Model at different stages of operation using the video analytics-based belt conveyor system, as disclosed in Figures 1-4.
Figures 6A-6F exemplarily illustrate charts that show actual operation scenario, application output, result (whether it’s true or false), accuracy of model, and confidence level, during operation using the video analytics-based belt conveyor system.
DESCRIPTION OF THE INVENTION
Exemplary embodiments now will be described. The disclosure may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey its scope to those skilled in the art. The terminology used in the detailed description of the particular exemplary embodiments illustrated in the accompanying drawings is not intended to be limiting. In the drawings, like numbers refer to like elements.
It is to be noted, however, that the reference numerals used herein illustrate only typical embodiments of the present subject matter, and are therefore, not to be considered for

limiting of its scope, for the subject matter may admit to other equally effective embodiments.
The specification may refer to “an”, “one” or “some” embodiment(s) in several lo cations. This does not necessarily imply that each such reference is to the same embodiment(s), or that the feature only applies to a single embodiment. Single features of different embodiments may also be combined to provide other embodiments.
As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless expressly stated otherwise. It will be further understood that the terms “includes”, “comprises”, “including” and/or “comprising” when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being “connected” or “coupled” to another element, it can be directly connected or coupled to the other element or intervening elements may be present. Furthermore, “connected” or “coupled” as used herein may include operatively connected or coupled. As used herein, the term “and/or” includes any and all combinations and arrangements of one or more of the associated listed items.
Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
The aim of the present invention is to provide a video analytics and deep learning-based application that enables remote monitoring of belt conveyor operation, more specifically a video analytics system that uses Artificial Intelligence and Machine Learning (AIML) tools to improve the surveillance and quick response to any deviations with respect to normal running conditions.

Referring to Figures 1A and 1B, Figure 1A exemplarily illustrates a basic flow diagram of the operation of the video analytics-based belt conveyor system 100, as an example embodiment of the present invention and Figure 1B exemplarily illustrates a phase 1 architecture of the video analytics-based belt conveyor system 100 operation. The application installed in the server 110 scans different frames 204 from the video data that is available from the camera feed 202 and retrieves frames 204 from the live broadcast of the video data. These pictures or frames 204 are fed into a programme or a model 112 that predicts anomalies or anomalies. As discussed before, the model 112 checks 206 for any anomaly, and if there is ‘no’ anomaly detected then the feed is continuously read 208. However, if an anomaly is detected (Yes) then the algorithm alerts 210 until the existence of the anomaly and the processor 108 proceeds with a subsequent command to fix 212 the anomaly. In the absence (no) of any anomalies, the subsequent frames are continuously retrieved from the cameras, for example, IP cameras. Based on the alert generated by the hardware processor 108, the anomaly needs to be fixed and if there is no alert, then the process is repeated 214 to check for further anomalies and accordingly alerts are raised.
As used herein, the term “anomaly” refers to any abnormality in the operation of the conveyor belt 114 that is recorded using the camera 102. As an example, Figures 5A-5E show different types of anomalies that may occur during operation of the conveyor belt 114. In other words, the surveillance camera 102 is placed in the proximity of the conveyor belt 114 to continuously transmit real time feed of video data via a real time stream protocol. As shown in Figures 2 and 3, the video feed is converted into frames and then the converted frames are passed to a model 112. The model 112 checks for any anomaly, and if there is no anomaly detected then the feed is continuously read. However, if an anomaly is detected then the algorithm alerts until the existence of the anomaly.
Referring to Figures 1B-1F, Figure 1B exemplarily illustrates a phase 1 architecture of the video analytics-based belt conveyor system 100 operation. Figure 1C exemplarily illustrates an overall architecture for multiple cameras in the operation of the video analytics-based belt conveyor system 100. Figure 1D exemplarily illustrates a camera configuration table for the operation of the video analytics-based belt conveyor system 100. Figure 1E exemplarily illustrates a camera transaction table for the operation of the video analytics-based belt conveyor system 100. Figure 1F exemplarily illustrates a model processing table for the video analytics-based belt conveyor operation system 100.

As shown in Figure 1B that shows the phase 1 architecture of the operation related to the video analytics-based belt conveyor system. A single camera 102 architecture is depicted in Figure 1B, which is employed check for anomalies or anomalies. The camera 102 is placed in the proximity of a conveyor belt 114 (as shown in Figure 1B) to continuously transmit real time feed of video data to the computing device 104 that comprises the memory storage 106 and the hardware processor 108. The memory storage 106 comprises stored instructions that when executed cause the hardware processor 108 to perform multiple operations. The hardware processor 108 converts the video data into multiple frames and the model 112 that is stored in the memory storage 106 of the computing device 104 compares the frames and checks for anomalies in working of the conveyor belt 114 based on characteristics of each frame. In response to an identified anomaly, information on the identified anomaly is identified by the hardware processor 108 of the computing device 106 and the hardware processor 108 generates an alarm based on the identified anomaly to alert a user regarding the working of the conveyor belt 114. Or in other words, the processor decides 216 to provide an alarm via a speaker 116, as shown in Figure 1B.
As shown in Figure 1B, the camera 102 transmits live feed of video data to the first computing device 102 that includes hardware processor 108 and the memory storage 106 (or the database). The hardware processor 108 or the CPU includes instructions in the form of functional instructions that are configured to be activated sequentially (Function 1, 2.., 13, …’n’ as shown in Figure 1B). The application installed in the computing device 102 searches for anomalies by analysing the frames from the video data that is retrieved from the camera 102. Considering that this is a sequential process, each abnormality is checked one by one and real-time updates to the memory storage 106, for example, a database, is made after the anomaly is found. The anomalies are chosen from the memory storage 106 by a decision-making component 216, which decides what to do with daily alerts and logs.
As shown in Figure 1C that shows the overall architecture for multiple cameras 102 in the video analytics-based belt conveyor system 100 operation. There are three elements to this current structure. The frames are retrieved from video data from each camera 102 and this information is saved on the memory storage 106, for example, hard disc. Secondly, the hardware processor 108 uses a different programme to analyse each frame of the video data individually to detect any irregularities or anomalies. Following prediction, the memory

storage 106 is updated. The hardware processor 108 then uses a third application to pull information from the memory storage 106, and decides how many alerts to be send, and decide how often to push logs. On the first computing device 104a with a GPU, the first two components will function (NVIDIA Quadro 2000). The first computing device 104a shares the processed data with the database 106. The second computing device 104b runs on Linux, which retrieves information from a shared database 106, and use its CPU or processor 108b to make decisions.
Figure 1D exemplarily illustrates a camera configuration table for the operation of the video analytics-based belt conveyor system 100. The name of the organisation will be showed in the first column of this table. The camera's IP and number are showed in the second and third columns. The latest pickle file that we'll utilise to find the anomaly is showed in the fifth column, and the fourth column displays the code for the anomaly. The type of function that will be used to create the anomaly will be showed in the sixth column, which will be the last. Referring to Figures 1E and 1F, as shown in Figure 1E that shows a camera transaction table for the operation of the video analytics-based belt conveyor system 100. After reading the frames from the cameras, the first programme will generate Table in Figure 1E. Here, the first column will display the video clip number, the second column the camera number, and the third column the video clip's duration. The name of the video file will appear in the fifth column, and the status of the video clip will appear in the final column (Initiated or Completed). Figure 1F exemplarily illustrates a model processing table for the operation of the video analytics-based belt conveyor system 100. This table is updated by a Program 2. The first column of this table will display the total number of video clips, and the second column will display their names. The video clip's state will be displayed in the third column, and the time-stamp of when it was processed will be displayed in the fourth column. The anomaly will then be displayed in the final column. The final program will read this table from the shared database and generates the alerts.
Figure 2 exemplarily illustrates an application workflow of the video analytics-based belt conveyor system 100 operation. Developing any model requires a series of steps which are as follows: Step 1: Gathering Data, Step 2: Preparing that Data, Step 3: Choosing a Model, Step 4: Training, Step 5: Testing, Step 6: Hyper parameter tuning, Step 7: Prediction, and Step 8: Deploy. The steps 402-422 are performed by the hardware processor. The video data is collected 402 from real time feed of camera throughout, for example, 24/7, and the extracted

404 frames from the input video data are saved into a folder and the data associated is labelled 406 into different classes. The labelling of the data is performed using a labelling tool.
Thereafter, pre-processing 408 steps are performed that comprise cropping frames, converting into grey scale, and scaling frames, etc. The cropping is performed on the frames so that the video data is divided into test and train parts. Furthermore, the model is trained 410 using the data that is pre-processed and this data is saved 414 in the memory storage. While training the model, hyper parameter tuning 412 is performed to extract the best parameters. The parameters include 1. Number of epochs, 2. Learning rate, 3. Training batch size, 4. Optimizer, and 5. Step_Lr_Gamma, etc. After training the model, testing 416 of the model is performed to facilitate better accuracy of the model in a real-world data. The model is then deployed 418 on a server using Flask Application Programming Interface (API), the prediction 420 of the frames is done using API in real time, and an alert 422 is generated based on model input.
Figure 3 exemplarily illustrates a Convolutional Neural Network (CNN) based deep learning algorithm 500 of the video analytics-based belt conveyor operation, as disclosed in Figures 1-4. The CNN is a type of deep learning model which deals with the frames associated with the video data in this particular case study. CNN is mathematical construct that is typically composed of three types of layers, namely convolution, pooling, fully connected (softmax). The first two layers i.e., convolution and pooling layer, perform feature extraction, whereas the third, a fully connected layer, maps the extracted features into final output. The term “softmax” as referred to herein, is also known as softargmax or a normalized exponential function, which converts a vector of K real numbers into a probability distribution of K possible outcomes.
Here, the input layer 502 receives the frames and normalizes 504 each frame so that the video data can be read by CNN algorithm. The model is defined 506 using the normalized frames and the first convolution later 508 applies a filter and creates a feature map. The first pooling layer 510 performs a reduction and the dimensions of the normalized frames and the second convolution layer 512 applies filters and creates another feature map. A second pooling layer 514 then performs another dimension reduction and a third convolution layer 516 applies

filters and creates another feature map. Finally, the softmax 518 chooses the maximum output, which is associated with the alert that needs to be generated based on the frames.
Figure 4 exemplarily illustrates an Object Detection Model of the video analytics-based belt conveyor system 100. To detect a human, the object detection model needs to be trained accordingly. Here, object will be human and all the data is collected 602 which consist of human and the data is annotated 604, i.e., drawn using bounding boxes around human in the frames. The training set is prepared by parsing 606 the annotated data and then a pre-trained model is generated accordingly and downloaded. The unknown dataset or the pretrained model 608 associated with new video data is used to train 610 the model and further the model is tested 612. The new data is imported 614 based on the tested model and the object, for example, human, is detected 616. Furthermore, the model is trained, and the best parameters are found using hyper parameter tuning. The new video data is imported to test the model in real world data to find the accuracy of model.
The data is classified into 4 different classes, namely Class1: Running with Coal, Class2: Running without Coal, Class3: Stop with Coal, and Class4: Stop without Coal. During classification of the frames, it is very difficult to find whether the conveyor is running or not since the frames are still and there is not much difference to detect conveyor is running or not. Furthermore, there is low accuracy for finding whether the conveyor is running or not, which is a barrier for creating good model with good accuracy. For this reason, illumination has been enhanced by providing a suitable light source above the conveyor. The light source is connected with the same power supply as the conveyor. As the conveyor starts running, the light source starts glowing which indicates that the conveyor is in motion, which increases the model accuracy for all the classes with great extent. Different programming approaches are used comprising compressing of frames, data augmentation, parallel processing, different algorithms, decoupling the application on client and cloud instances, etc., to overcome all the issues and increase the accuracy.
Figures 5A-5E exemplarily illustrate an Object Detection Model at different stages using the video analytics-based belt conveyor operation, as disclosed in Figures 1-4. Figure 5A shows the conveyor unit running in idle check. Figure 5B shows the conveyor unit in stop and loaded check. Figure 5C shows the conveyor unit in normal running check. Figure 5D shows the conveyor unit in normal stop check. Figure 5E shows the conveyor unit in miss-align

check. Figures 6A-6F exemplarily illustrate charts that illustrate actual operation scenario, application output, result (whether it’s true or false), accuracy of model, and confidence level.
The video analytics-based belt conveyor operation system provides a unique kind of supervision and control developed for belt conveyor operation, which requires human brain-like processing, decision making and outcome, and do not require PLC based controls using limit switches, proximity switches, rpm sensors, belt weigher inputs, etc. With help of single frame various functionalities are monitored at pre-defined interval and the video analytics-based belt conveyor operation system gives voice output to control room engineers as, for example, alert message, SMS and mails, phone calls, record output data, etc. The benefits of video analytics-based belt conveyor operation system comprise cost saving through manpower optimization in conveyor plant, improved supervision and quick response, auxiliary power saving, reduced R&M cost, improved equipment availability, automation, etc.
In the specification, there has been disclosed exemplary embodiments of the invention. Although specific terms are employed, they are used in a generic and descriptive sense only and not for purposes of limitation of the scope of the invention. Although the invention has been described with reference to specific embodiments, this description is not meant to be construed in a limiting sense. Various modifications of the disclosed embodiments, as well as alternate embodiments of the invention, will become apparent to persons skilled in the art upon reference to the description of the invention. It is therefore, contemplated that such modifications can be made without departing from the spirit or scope of the present invention as defined.

We Claim:
1. A video analytics-based belt conveyor system comprising:
a camera placed in a proximity of a conveyor belt to continuously transmit real time feed of video data to a computing device that comprises a memory storage and a hardware processor;
the memory storage comprises stored instructions that when executed cause the hardware processor to perform a plurality of operations;
the hardware processor converts the video data into plurality of frames;
a model that is generated based on a machine learning method and stored in the memory storage of the computing device compares the plurality of frames and checks for anomalies in working of the conveyor belt based on characteristics of each frame;
in response to an identified anomaly, information on the identified anomaly is detected by the hardware processor of the computing device; and
the hardware processor generates an alarm based on the identified anomaly to alert a user regarding the working of the conveyor belt.
2. The video analytics-based belt conveyor system as claimed in claim 1, wherein if there is no anomaly detected during the anomaly check, then the hardware processor continuously evaluates real time feed of subsequent video data.
3. The video analytics-based belt conveyor system as claimed in claim 1, wherein a plurality of the cameras are positioned in the proximity of a plurality of the conveyor belts to continuously transmit the real time feed of the video data to the computing device, wherein the hardware processor retrieves the frames from the video data from each camera and analyses each frame of the video data separately to identify the anomalies, and determines number of the alerts to be generated based on the anomalies.
4. The video analytics-based belt conveyor system as claimed in claim 1, wherein the hardware processor is configured to:
collect real time feed of the video data from the camera;
extract the frames from the video data, which is saved into a folder;
label the frames from the video data into different classes using a labelling tool;

pre-process the labelled data, wherein the pre-processing comprises cropping the frames, converting the frames into grey scale, and scaling the frames;
train the model using the pre-processed data and save the trained data in the memory storage;
hyper parameter tune the trained data to extract the best parameters;
test the model to facilitate accuracy of the model in a real-world data;
deploy the model on a server using an interface;
predict the frames using the interface in real time; and
generate an alert.
5. The video analytics-based belt conveyor system as claimed in claim 1, wherein the
machine learning method is a Convolutional Neural Network (CNN) based deep learning
algorithm, which comprises steps to:
receive, via an input layer, the frames from the video data;
normalize each frame so that video data is readable by the CNN algorithm;
define the model using the normalized frames;
apply filters using a first convolution layer to create a feature map;
perform, via a first pooling layer, a reduction in dimensions of the normalized frames;
apply filters using a second convolution layer to create another feature map;
perform, via a second pooling layer, another reduction in dimensions of the normalized frames;
apply filters using a third convolution layer to create another feature map; and
choose a maximum output, via SoftMax, in the form of the alarm that needs to be generated based on the frames.
6. A method of detecting anomalies in a belt conveyor operation comprising:
providing a camera that is placed in a proximity of a conveyor belt;
continuously transmitting real time feed of video data to a computing device that comprises a memory storage and a hardware processor, wherein the memory storage comprises stored instructions that when executed cause the hardware processor to perform a plurality of operations that includes:
converting the video data into plurality of frames;

comparing, via a model that is generated based on a machine learning method, the plurality of frames and checking for anomalies in working of the conveyor belt based on characteristics of each frame;
in response to an identified anomaly, detecting information on the identified anomaly; and
generating an alarm based on the identified anomaly to alert a user regarding the working of the conveyor belt.