Integrating Artificial Intelligence and Deep Learning for the Purpose of Detecting Arms in Defense

Abstract

There has been an increase in the relevance of security as a primary issue across all sectors as a result of the prominence of criminal activities that take place at public events or in settings that seem to be distant. It is important to note that computer vision plays a vital role in the fields of anomaly detection and monitoring. It provides a broad range of applications that may be used to handle a variety of issues. The technology of computer vision is used in several applications. An essential component of the intelligence monitoring process is the use of video surveillance systems that are equipped with the capacity to identify and assess the surrounding environment, in addition to identifying and analysing unexpected events. The explanation for this occurrence is that there is a rising need to protect personal assets, and there is also an increased focus on ensuring that there is an increased degree of safety and security. This study makes use of a Yolo algorithm in order to effectively accomplish the goal of automating the identification of guns and other types of weapons. The method that is being presented involves the use of two distinct types of datasets. The first dataset is made up of images that have already been classified, whereas the second dataset is made up of photographs that have been manually labelled. The use of these datasets is available to the public. Despite the fact that the laborious tabulation of data is carried out with accuracy, the actual use of these findings in real-world circumstances may be contingent on the delicate balance that is established between speed and precision.

Introduction

I. INTRODUCTION

The process of recognizing occurrences or things that are irregular, unexpected, unpredictable, and uncommon is referred to as weapon or anomaly detection. These occurrences or items are deemed to be non-routine occurrences or irregular things that depart significantly from the patterns or items that are already included in a collection. In the context of pattern analysis, the word "anomaly" refers to a pattern that is not consistent with the generally accepted set of regular patterns. In the particular phenomenon that is being investigated, the incidence of anomalies is dependent on the unique phenomenon. The term "object detection" refers to a computational method that uses learning algorithms or models to recognize and categorize instances of different types of things, all while gathering pertinent information from these objects. One of the most important aspects of the implementation that is advised is the precise identification and categorization of guns. The possibility of unfavorable reactions that may follow from a misleading warning is the impetus behind the need to address accuracy.

It is necessary to strike a balance between speed and precision in order to successfully choose the method that is most suited to the situation. During the process of input, frames are extracted from the video that is being used. Following the execution of the frame differencing procedure, which ultimately results in the formation of a bounding box, the identification of an item is carried out afterwards. Immediately after the creation of the dataset, the object detection algorithm is put through training using the dataset in question. In order to determine which method would be most appropriate for gun detection, it was necessary to take into account the particular application at hand. SSD and quick RCNN were the two possibilities that were seriously investigated.

For the purpose of addressing the issue of detection, this technique makes use of a number of different machine learning models, including the region convolution neural network (RCNN) and single-shot detection.

A. Model Diagram

B. Problem Statement

In a number of places, particularly in nations where the ownership and use of such weapons is legal under certain conditions, the frequency of criminal activity involving the use of knives or firearms has dramatically increased. By analyzing surveillance video, it is possible to identify the presence of potentially hazardous items, such as knives and firearms, with the intention of reducing the number of events that occur as a result of such occurrences. In order for the current surveillance and control system to work well, it is still necessary for human monitoring and interaction to be present. In this research, a system is shown that demonstrates the power to independently distinguish firearms in video footage. As a result, the system is suitable for application in situations involving surveillance and control. The purpose of this research is to argue for the use of deep learning methods in order to identify such weapons in a timely manner. Taking this technique would include the implementation of video monitoring in real time, with the intention of reducing the amount of processing work required.

II. LITERATURE REVIEW

Wei Liu et al., Conventionally The first material that is employed in the construction business is cement, which is a substance that is extensively used, acts as a major binder, and is an essential component in the production of concrete. For traditional cement to be manufactured, a high level of technical expertise is required, and a significant amount of energy is used throughout the production process. Over the course of the last ten years, the most common strategy has been to reduce the amount of cement that is used in the manufacture of concrete and to incorporate waste materials into the manufacturing process. It is well known in the academic literature that the manufacturing of traditional Portland cement is associated with a number of risks. It is only through the use of waste resources and the reduction of the emission of breathing gases into the atmosphere that it is possible to make the claim that they are environmentally friendly. Flour ash, in combination with fine aggregate and coarse aggregate, is the principal material that is used for a broad variety of processes. A large rise in the price of river sand has occurred as a consequence of a scarcity of this element, which is a crucial component. Concrete, on the other hand, has a vital role within the building sector. River sand, on the other hand, is another essential component. Sand is becoming more scarce as a result of the growing demand for this resource in the building industry, which has led to a significant problem. Casting geopolymer specimens does not include the integration of extra replacement materials, such as rock dust obtained from a quarry site, in terms of their concentration. This is because the method does not involve the incorporation of such materials. In spite of the fact that a number of research publications present alternative materials as partial substitutions rather than complete replacements, this strategy is still widely used. Researchers are focusing their attention on the assessment of test findings acquired from GPC specimens, which include steel fibers and glass fibers. This is the primary topic of interest for these researchers. The corpus of literature currently available indicates that there has been a glaring lack of research on natural and hybrid fibers.

D. Erhan et al., In recent years, deep convolutional neural networks have shown remarkable performance on a variety of image recognition benchmarks, the most notable of which is the ImageNet Large-Scale Visual Recognition Challenge (ILSVRC-2012). The advancements that have been made in neural networks have made the aforementioned feat much easier to achieve. The network that was able to achieve success in the localization subtask was the one that was able to predict a single bounding box and a confidence score for each kind of item that was shown in the picture.

It is possible for this specific model to grasp the overall context of the picture that encompasses the things on the screen. On the other hand, it is unable to efficiently manage many occurrences of the same object inside the picture, and instead, it uses a straightforward method that involves duplicating the number of outputs for each instance. In this paper, a neural network model for detection is presented. The model makes use of saliency as its source of inspiration. The model that has been provided is capable of making predictions on a collection of bounding boxes that are not particular to any particular class. It assigns a single score to each box, which represents the chance that the box contains an item of interest. It is possible to generalize across classes at the highest layers of the neural network thanks to the model's capacity to efficiently manage a variable number of examples for each class in a seamless way. The model demonstrates this capability. Based on the VOC2007 and ILSVRC2012 datasets, we are able to achieve recognition performance that is comparable to that of our rivals. This is accomplished by focusing on just the highest-ranked predicted sites in each picture and doing a limited number of evaluations using neural networks. Utilizing just the regions inside each photo that have the greatest rating is the method that is used to accomplish this.

III. PROPOSED METHODOLOGY

When it comes to the implementation that is advised, the identification and classification of guns are given the highest priority. The possibility of unfavorable reactions that may follow from a misleading warning is the impetus behind the need to address accuracy. Finding the best way to achieve the desired balance between speed and accuracy is essential, and the identification of the best strategy is essential. Within the framework of a Convolutional Neural Network (CNN)-based Single Shot MultiBox Detector (SSD) and a Faster Region Convolutional Neural Network (Faster RCNN), we proposed a method for the detection of weapons. There are two different kinds of datasets that are used in the strategy that has been presented. One dataset is made up of photographs that have already been recognized, whereas the other dataset is made up of photographs that have been manually categorized. Both of the algorithms acquire a significant level of accuracy while they are working through the process of computing the results. On the other hand, the usefulness of these algorithms in real-world scenarios can be contingent on the manner in which they strike a compromise between accuracy and efficiency.

A. Frame Extraction

Frame extraction is a powerful way to include video content because it permits the selection of a condensed collection of significant frames that effectively represent the core of video sequences. This makes frame extraction a strong method for incorporating video information. The material can be implemented more easily as a result of this situation. Due to the fact that they do not fulfill specific requirements, the majority of the key frame extraction methods that are now available are considered to be insufficient for the purpose of protecting video copyright.

B. Boundary Tracking

Identifying the pixels that are located on the border of a binary digital area is the objective of the segmentation method known as boundary tracing, which is also widely referred to as contour tracing. According to one interpretation, this technique might be seen as a methodology that defines the boundaries of the digital realm. When it comes to the investigation of the aforementioned territory, the first step that is regarded as crucial and essential is the process of investigating the boundaries.

A contour is the delineation of the boundary of an image, in contrast to an image edge, which is often described as a sudden shift in low-level visual qualities such as brightness and color. A contour is a reference to the demarcation of the border of an image. On the other hand, a contour may be used to define an image edge. This latter term refers to the process by which ownership of a pixel is transferred from the surface of one entity to the surface of another person or thing.

C. Frame Differencing

The computer performs a computational operation known as frame differencing, which involves determining the difference between two successive video frames. This is done by calculating the difference between the frames. In the event that the pixels have been altered, this would be indicative of an alteration included within the picture, maybe suggesting the existence of motion. The vast majority of solutions often make use of blur and threshold approaches in order to differentiate between noise and true movement.

D. Detection Objects

The field of computer vision has a specific subfield known as object detection. This subfield is responsible for enabling computational systems to recognize, identify, and categorize visually observable items that are shown in pictures or videos. There is a possibility that the utilization of computer vision methods will make the accomplishment of this purpose easier. An image is a single frame that records a single instance of a naturally occurring event in its stationary form. This is the definition of an image. For the goal of identifying, localizing, and categorizing objects in photographs and videos, a number of specialized algorithms have been created; these algorithms have been developed. RCNNs, which stand for region-based convolutional neural networks; SSDs, which stand for single-shot multibox detectors; RetinaNet; and YOLO, which stands for you only look once, are some of the approaches that fall under this category. Many algorithms have been developed specifically for the purpose of carrying out certain activities.

IV. TESTING STRATEGY

This In the context of software testing, the process of running a software program with the intention of locating and diagnosing faults is often referred to as software testing. We must ensure that our program is free of any errors in order to achieve the highest possible level of functioning. In the event that the testing procedure is carried out effectively, all of the software's current issues will be resolved.

A. Types of Testing

1. White Box Test: The logic-based testing technique is one of the testing approaches that is dependent on having an understanding of the underlying logic that is contained inside the code of an application. This method incorporates a number of checks, such as coverage of code statements, coverage of branching, coverage of routes, and coverage of conditions. The term "software developer" is often used to refer to the person who is accountable for carrying out the responsibilities associated with software development.
2. Black Box Testing: Functional regression testing is a technique that evaluates the functioning of an application without needing detailed knowledge of the program's code or internal structure. This enables the assessment to be performed without any restrictions. In the process of doing the assessments, both the functioning and the needs are taken into account.

Table No .1 Black box Testing

3. Unit Testing: When it comes to software verification and validation, the process involves a programmer examining the individual units of source code to determine whether or not they are suitable for usage. Generally speaking, the development team is the one that takes on the responsibility of moving the project forward through its many stages.

4. Integration Testing: The process of testing software includes a phase in which separate software components are merged and then tested as a united entity. This phase is known as the integration phase. In most cases, the testing teams are the ones that are accountable for carrying out the tests.

5. Alpha Testing: The kind of testing that is carried out on a software product or system at the location where the developer is also located typically, the end users themselves carry out the execution.

B. Beta Testing

It is necessary for the application to go through one further round of testing before it can be made available for use in business settings. End users or other workers are often the ones who carry out the job.

Table 2: Test Cases

V. RESULTS

Figure 7.3.1 provides a visual representation of the application known as the anaconda command prompt, which is used for the purpose of carrying out code execution.

Conclusion

An increasing focus on security has been placed across all industries as a result of the surge in criminal behavior that has been seen during large gatherings or in regions that are isolated. Within the field of intelligence monitoring, the use of video surveillance systems that are able to detect and analyze the scene, in addition to recognizing odd occurrences, is of important value. It is the underlying cause of this phenomenon that there is a growing need for the protection of personal property, the guarantee of personal safety, and the maintenance of security. This research intends to identify weapons via the use of recursive convolution neural networks (RCNN), which are a kind of neural network. During the training phase, graphics processing units (GPUs), high-end digital signal processing (DSP), and field-programmable gate array (FPGA) packages may be used to analyze larger datasets. This allows for more efficient performance

References

[1] Wei Liu et al., “SSD: Single Shot MultiBox Detector”, European Conference on Computer Vision, Volume 169, pp 20-31 Sep. 2017. [2] Erhan et al., “Scalable Object Detection Using Deep Neural Networks,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR),2014. [3] Ruben J Franklin et.al., “Anomaly Detection in Videos for Video Surveillance Applications Using Neural Networks,” International Conference on Inventive Systems and Control,2020. [4] H R Rohit et.al., “A Review of Artificial Intelligence Methods for Data Science and Data Analytics: Applications and Research Challenges,”2018 2nd International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud), 2018. [5] Abhiraj Biswas et. al., “Classification of Objects in Video Records using Neural Network Framework,” International conference on Smart Systems and Inventive Technology,2018. [6] Pallavi Raj et. al.,“Simulation and Performance Analysis of Feature Extraction and Matching Algorithms for Image Processing Applications” IEEE International Conference on Intelligent Sustainable Systems,2019. [7] Mohana et.al., “Simulation of Object Detection Algorithms for Video Surveillance Applications”, International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud),2018. [7] Yojan Chitkara et. al.,“Background Modelling techniques for foreground detection and Tracking using Gaussian Mixture model” International Conference on Computing Methodologies and Communication,2019. [8] Rubner et.al, “A metric for distributions with applications to image databases”, International Conference on Computer Vision,2016. [9] N. Jain et.al., “Performance Analysis of Object Detection and Tracking Algorithms for Traffic Surveillance Applications using Neural Networks,” 2019 Third International conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud), 2019. [10] A Glowacz et.al., “Visual Detection of Knives in Security Applications using Active Appearance Model”,Multimedia Tools Applications, 2015. [11] S. Pankanti et.al.,“Robust abandoned object detection using regionlevel [12] analysis,”International Conference on Image Processing,2011. [13] Ayush Jain et.al.,“Survey on Edge Computing - Key Technology in Retail Industry” International Conference on Intelligent Computing and Control Systems,2019. [13] Mohana et.al., Performance Evaluation of Background Modeling Methods for Object Detection and Tracking,” International Conference on Inventive Systems and Control,2020. [14] J. Wang et.al., “Detecting static objects in busy scenes”, Technical Report TR99- 1730, Department of Computer Science, Cornell University, 2014.

Copyright

Copyright © 2023 Seelam SKML Sandhya, M Pavani Devi, V Anil Santosh, D D D Suribabu. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.