Description:Field of Invention
The innovation relates to the use of artificial intelligence-based deep learning algorithms to identify people carrying weapons in public places in order to improve or offer security for people and prevent injury.
The Objectives of this Invention
The main objective of the innovation is to develop and implement a reliable method for locating the weapon in the immediate vicinity. A video that contains weapons is the input. Applying the Faster RCNN technique, the different sections of the weapon are identified as part of the computation. Weapons are going to be recognized if any are found following identification. The output identifies the specific kind of weapon one is found.
Background of the Invention
In recent years, every Government/private firm lacking to identifying the weapon detection in a public places. First, type of technique has been introduced in (US2020/10964177B1), A gunshot detection/security system utilized by schools or other buildings has one or more pods scattered throughout the building's grounds. Each pod has a camera, a thermal camera, and an acoustic sensor for detecting video, pictures, heat signatures, and sound inside a detection region for the specific pod. The information collected by the sensors is then analysed to find a risky occurrence in the building, and notifications about the potentially hazardous occurrence are sent to students or other building inhabitants, executives, parents, and response personnel via pods or client computing devices. Digital visualisations of the building's inside and outside are created by a server computer device with location data on the building's inhabitants and a danger zone marking the epicenter of the harmful occurrence. Another method, (US2021/0209402A1), Weapons are found and tracked using a procedure. A camera sends a frame to create a video. Employing a weapon recognition approach, a weapon was found in the frame. A weapon match classifier is used to categorise the weapon of choice from the picture. The classification of the weapon results in the production of a weapon alert. In reaction to the weapon alarm, the video is shown. In (US2021/0382166A1), A weapon recognition system with a radar component and a magnetometer is part of certain apparatuses. A series of radio frequency (RF) reaction waves from an object under test (IUT) are arranged to be detected by the radar device. A magnetic reaction signals from the IUT are set up to be detected by the magnetometer. The weapon detection system uses a set of RF response signals and magnetic response signals of the set to create an amalgamated multi-source recognition indication. In (US2019/11275925B2), The technologies described here can be used to detect past, present, or imminent assault using computer vision. A real-time human behaviour recognition system with object categorization is one of the innovations that can be utilised as an intelligent addition to surveillance security equipment. The innovations can be applied to unmanned aeroplanes, monitoring systems, and security cameras. The algorithms can improve their ability to identify violence by utilising several types of machine learning. Additionally, these methods are of combining precise results.
The (Jain et al [2020], 2020 International Conference on Electronics and Sustainable Communication Systems (ICESC), Coimbatore, India, 2020, pp. 193-198), This study uses the SSD and Faster RCNN convolution neural network (CNN) methods to accomplish automatic gun (or weapon) identification. The suggested implementation employs two different datasets. One dataset had already labeled photographs, and the other contained images that needed to be individually classified. Both algorithms produce excellent precision in tabulated outcomes, but their practical use may depend on the compromise between time and precision.
The (Yadav et al [2023], Expert Systems with Applications, Volume 212, 2023, 118698), We have outlined in this study the weaknesses of the weapon detecting systems that are currently available. The computerized identification of guns and other weapons might aid investigations into crime scenes. Identification of the precise firearm used in an attack, also known as intra-class identification, is a novel and challenging area of investigation. The study uses traditional machine learning and deep learning methodologies to analyze and categorize the benefits and drawbacks of many existing algorithms used to detect various types of weapons.
Summary of the Invention
The already labeled video collections are used to model the Faster RCNN method. Both techniques are efficient and produce superior outcomes, but employing them in real-time necessitates a speed-accuracy trade-off. More slowly than SSD, faster RCNN yields an average frame rate of 1.606 seconds per frame. Faster RCNN offers an accuracy rating of 84.6 percent. SSD's accuracy is 73.8% lower than the quicker RCNN's. The more precise prediction was achieved with faster RCNN.
Detailed Description of the Invention
The proposed invention has the following two stage process; such as
The proposed system focuses on the introduction of certain realistic AI-based weapon identification algorithms that may be used to enhance current standard methodologies. It uses the SSD and Faster RCNN convolutional neural network (CNN) strategies for weapon recognition. The earlier model can now match the Faster R-CNN's precision while using fewer focused images, increasing efficiency and cutting costs.
After Fast R CNN, the R-object CNN detection network design, was upgraded. Despite the network's development and recognizing times being greatly sped up, it still needs to be faster to be used as a real-time system because it takes around two seconds to generate an input image. An technological bottleneck is a form of focused search. As a result, K He et al. created the Faster R-CNN framework. They offer a different method for producing suggestions for regions called the Region Proposal Network in place of a focused search. The design includes the Region Proposal Network (RPN) and the Object Detection Network.
YOLO is a Convolutional Neural Network (CNN) for real-time object identification. The classifier-based systems known as CNNs are capable of identifying patterns in incoming images as organised arrays of data. The advantage of YOLO is that it is significantly faster than conventional systems while maintaining accuracy. To inform the projections it makes about the overall picture of the image, it stimulates the model's abilities to evaluate the complete image throughout the period of testing. YOLO and other CNN algorithms assign a "score" to regions based on their resemblance to predefined classifications.
The process flow of weapon detection using RCNN (Region-based Convolutional Neural Network) and artificial techniques can be divided into several steps. Here's a high-level overview of the process:

Dataset Preparation: Gather a large dataset of images containing both weapon and non-weapon objects. The dataset should be diverse and representative of real-world scenarios. Annotation: Annotate the images in the dataset by labeling the regions or bounding boxes that contain weapons. This step helps train the RCNN model to learn the visual features of weapons. Training the RCNN Model: Utilize the annotated dataset to train the RCNN model. The model consists of two components: a region proposal network (RPN) and a classifier. The RPN proposes potential regions of interest (ROIs) in the image, and the classifier categorizes each ROI as a weapon or non-weapon. Feature Extraction: Extract visual features from the ROIs proposed by the RPN. These features capture the discriminative information necessary for distinguishing weapons from non-weapons. Training the Classifier: Train the classifier using the extracted features and their corresponding labels (weapon or non-weapon). This step involves optimizing the model parameters using techniques like backpropagation and gradient descent. Fine-Tuning: Fine-tune the trained model to improve its performance. This process involves adjusting the model's hyperparameters, such as learning rate, regularization, and network architecture, to enhance detection accuracy. Testing: Evaluate the trained model on a separate test dataset. The model takes an input image, generates region proposals using the RPN, extracts feature from the proposed regions, and classifies them as weapons or non-weapons using the trained classifier. Post-processing: Apply post-processing techniques to refine the detection results. This step may include removing duplicate detections, filtering out false positives, and adjusting the bounding boxes to tightly fit the detected weapons. Deployment: Deploy the trained model in a real-world scenario for weapon detection. The deployment may involve integrating the model into a surveillance system, video processing pipeline, or any other application where weapon detection is required.
It's worth noting that artificial techniques, such as data augmentation, transfer learning, and ensembling, can be employed at various stages of the process to improve the performance and generalization capabilities of the weapon detection system. Additionally, the specific implementation details and choices of algorithms may vary depending on the particular RCNN variant used, such as Faster R-CNN or Mask R-CNN.
We independently tested three object identification methods for modeling: Faster RCNN, YOLO, and RCNN. Sliding window-based object identification techniques were abandoned since studies indicated they were ineffective. Additionally, research indicates that Faster RCNN has a higher recall than YOLO. Yolo is really quick for real-time detection, though. Additionally, the rate of false positive detection is relatively high for well-detailed photos. To combat this, we improved our classification model with new data and a progressive unfreezing of the top layers. In order to lower false positives, the erroneously discovered photos are incorporated into the categorization model. The pre-labeled video datasets are used to model the Faster RCNN method. Both techniques are efficient and produce superior outcomes, but employing them in real-time necessitates a speed-accuracy trade-off. Slower than SSD, faster RCNN yields a frame time of 1.606 seconds per frame. Faster RCNN offers an accuracy of 84.6 percent. SSD's accuracy is 73.8% lower than the quicker RCNN's. Faster RCNN produced higher accuracy.
Weapon detection using artificial intelligence has various applications across different domains. Here are some of the notable applications:

Public Safety and Security: Weapon detection systems can be deployed in public spaces, such as airports, train stations, schools, and stadiums, to enhance security measures. These systems can help identify concealed weapons, such as guns or knives, and alert security personnel in real-time. Border Security: Weapon detection AI can be utilized at border checkpoints and ports of entry to identify illegal firearms, explosives, or other weapons being smuggled across borders. This enhances the effectiveness of border security measures and helps prevent illicit activities. Military Applications: In military settings, weapon detection AI can assist in identifying and tracking armed individuals or vehicles, helping to enhance situational awareness and improve decision-making in combat scenarios.
Workplace Security: Weapon detection AI systems can be utilized in workplace environments to enhance employee safety and prevent incidents of violence. By continuously monitoring areas or entrances, the technology can quickly identify and alert authorities to the presence of weapons. Transportation Security: AI-based weapon detection can be integrated into transportation security systems, including airport security checkpoints and mass transit facilities. This enables the detection of concealed weapons, ensuring the safety of passengers and preventing potential threats to transportation infrastructure.
It's important to note that while AI-based weapon detection systems can be powerful tools for enhancing security, they should be used in conjunction with other security measures and human judgment for optimal effectiveness.
4 Claims & 1 Figure
Brief description of Drawing
In the figure which are illustrate exemplary embodiments of the invention.
Figure 1, The Process of Proposed Invention , Claims:The scope of the invention is defined by the following claims:

Claim:
1. A system/method to detect the weapons using the Artificial Intelligence based Deep Learning algorithms, said system/method comprising the steps of:
a) The system starts with datasets collection from various cameras (1), from that all the attributes given to make the datasets (2).
b) Then proposed invention is incorporated preprocessing steps (3), to identify some of the important predictable images (4), the filter data is feature extraction process (5), the image is matched and accuracy metric was compared in (6), then finally the weapon is predicted by the user (7).
2. As mentioned in claim 1, the invented system starts with various videos and image dataset uploading to start the process.
3. According to claim 1, the preprocessing will initiate to remove the noisy data from the dataset and it will trigger feature extraction process of RCNN algorithms to split the data into training and testing part.
4. According to claim 1, now, the proposed invention will start from RCNN functions, then this will be matched with captured figure and detects the weapons and type of weapons using RCNN architecture based deep learning algorithms.