Weapon Detection using Yolov4, CNN

Abstract

A Weapon Detection software is a very efficient way to monitor the streets and alert the operator only when there is a situation where people carrying a weapon like a gun or knife can be detected and analyzed with the help of Yolov4 object detection algorithm. And the notification of the same will be given via email to the CCTV operator with the photo where people carrying a gun or knife can be seen and also the location of the CCTV. So this will enhance the safety and security of any city with the requirement of limited man power. The proposed paper focuses on different approaches and algorithms that could potentially help in Weapon detection.

Introduction

I. INTRODUCTION

To monitor and notify only when there's a situation where a person or group of people carrying weapons like gun, rifle or even a knife can be detected and notified via email to the CCTV operator. Using YOLOV4 Object detection algorithm objects can be classified using pre trained custom files trained from google colab, it is free for the public and it can be used for machine learning applications. With the free cloud service and most importantly free GPU with limited hardware requirements we can custom train the files using google colab. CNN is used for Deep learning for object recognition. It achieves excellent object recognition accuracy by using a deep ConvNet to classify object proposals. R-CNN has the ability to scale to thousands of object classes without resorting to approximate techniques and including hashing. These custom trained files can be used to detect specific types of objects in our case, the weapons like guns, pistols, rifles and even a knife. Such objects can be detected and when in a CCTV footage such objects are detected then using the SMTP Library in python an email notification is sent to the CCTV Operator with details such as image of guns and the location of the CCTV camera.

II. LITERATURE SURVEY

III. ALGORITHMS

After reviewing the literature, we found that for weapon detection two main algorithms were used. In this section, we are discussing those algorithms.

A. Yolov4 Algorithm

YOLOv4 is a real-time object detection algorithm that stands for "you only look once" identifies specific objects in videos, live feeds, or images. YOLO uses features learned by a deep convolutional neural network for detecting an object. YOLOv4 is an upgraded version of YOLOv3.It is a real- time object recognition system that can recognize multiple objects in the single frame. YOLO is also known for its speed. As speed and accuracy are inversely proportional. Accuracy remains still in the area of development. But in YOLO v4 the accuracy rate has also been improved. It can predict up to 9000 classes and even unseen classes. The real-time recognition of system will recognize multiple objects from an image and also make a boundary box around the object. It can be easily trained and deployed in a production system. YOLO is based upon a single Convolutional Neural Network (CNN). The CNN divides an image into regions and then it predicts the boundary boxes and probabilities for each region. It simultaneously predicts for multiple bounding boxes and probabilities for those classes. YOLO sees the entire image during training and test time so it implicitly encodes for corresponding information about classes as well as their appearance.

YOLOv4 is almost twice as fast as EfficientDet (competitive recognition model) with comparable performance. In addition, AP (Average Precision) and FPS (Frames Per Second) increased by 10% and 12% compared to YOLOv3.

B. Harris Corner Detection Algorithm

Harris Corner Detector is a corner detection operator that is used to extract corners and infer features of an image. Corner can be interpreted as the junction of two edges, where an edge is sudden change in an image brightness.

The steps for detecting corners are

1. Converting the color image into grayscale
2. Calculation of spatial derivative
3. Setting up tensor structure
4. Calculation of Harris response
5. Non-Max suppression to consider only the closet one

It is a single point feature extraction algorithm. It is the difference in intensity of all the directions for a displacement of (u, v), can be shown in equation.

 E(u, v)= Exy w(x,y) [1(x + u,y+ v)-1(x,y)]² 1

Here, window function can be a gaussian window giving weights to pixels underneath or a window which is rectangular. For corner detection E (u, v) must be maximized and this can be done by applying Taylor Expansion which will give us final equation stated as equation 2 and 3.

E(u, v) ≈ [u_v] M[u] Where

M = {x,y w (x, y) Ixly lyly [Lxlx Ixly]

Here Ix and Iy are the derivatives of image in (x, y) and finally we measure the corner response as equation 4 R = det (M) - k (trace (M))²

Where det(M)=1^2 trace (M)=A1+X2

A1 and A2 are the eigen values of M k is a constant

C. CNN (Convolutional Neural Network)

CNN stands for "Convolutional Neural Network" and is largely utilized in image processing. The number of hidden layers employed between the input and output layers determines the CNN's strength. A set of features is extracted by each layer. A series of filters are applied to the input to create feature maps. Each filter multiplies its weights by the input values after going over the full input. The result is sent to a Rectified Linear Unit (ReLU), sigmoid, or the activation function. The set of weights is evaluated using a loss function. The feature maps that the filters produce highlight various aspects of the input.

Even though CNN is commonly utilized in image and video processing, recent approaches to NLP utilize CNN. A pre- processing step in NLP converts the text input to a matrix representation. Sentence characters are used as rows, and alphabet letters are used as columns, in the matrix form. In NLP, a filter is slid across the matrix's words. As a result, the words are detected using the sliding window technique.

Conclusion

YOLO is a futuristic recognizer that has faster FPS and is more accurate than available detectors. The detector can be upskilled and used on a conventional GPU which enables extensive adoption. New features in YOLOv4 improve accuracy of the classifier and detector and may be used for other research projects.

References

[1] Z.-Q. Zhao, P. Zheng, S.-T. Xu and X. Wu, ”Object Detection With Deep Learning: A Review,” IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, pp. 1-21, 2019 [2] S. Tej and D. Kumar Vishwakarma, ”Human Activity Recognition in Video Benchmarks: A Survey,” Advances in Signal Processing and Communication, pp. 247-259, 2019 [3] P. M. Kumar, U. Gandhi, R. Varatharajan, G. Manogaran, R. Jidhesh, and T. Vadivel, “Intelligent face recognition and navigation system using neural learning for smart security in internet of things,” Cluster Computing, vol. 22, no. S4, pp. 7733–7744, 2019 [4] V. Babanne, N. S. Mahajan, R. L. Sharma, and P. P. Gargate, “Machine learning based smart surveillance system,” in Proceedings of the 2019 0ird International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud)(ISMAC), pp. 84–86, IEEE, Palladam, India, December 20 [5] S. Minaeian, J. Liu and SonYoung- Jun, ”Effective and Efficient Detection of Moving Targets From a UAV’s Camera,” IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, vol. 19, no. 2, pp. 497-502, 2018. [6] F. Gelana and A. Yadav, ”Firearm Detection from Surveillance Cameras Using Image Processing and Machine Learning Techniques,” in Smart Innovations in Communications and Computational Sciences (ICSICCS 2018), 2018. [7] A. Castillo, S. Tabik, F. Perez, R. Olmos and F. Herrera, ”Brightness guided preprocessing for automatic cold steel weapon detection in surveillance videos with deep learning,” Neurocomputing, pp.151- 161, 2018.

Copyright

Copyright © 2022 Atharv Belurkar, Ashish Waghmare, Sahil Mallick, Nikhil Waghamode, Prof. Reshma Totare . 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.