A Comparative Approach for Weapon Detection from Images Using Deep Learning Algorithms
Y. V. Srinivasa Murthy, Smiti Agrawal, Reya Malu, Linga Reddy Cenkeramaddi, K.S. Madhusudhan · 2024
Recent reports claim that the global crime rate is on the rise. There has been a significant increase in gun violence and crimes. According to polls, the portable pistol is the most regularly used handgun in a range of crimes, such as break-ins, robberies, thefts, and assaults. These situations can be averted if misbehavior is detected early on and suspicious behaviors are monitored constantly so that law enforcement officers can respond appropriately. With the growth of AI and computer vision technologies, we made an effort to automatically detect the weapon using various deep learning approaches. The standard dataset provided by Kaggle with 6000 gun images and YOLO labels has been taken for experimentation. Three different deep learning architectures such as YOLOv3, faster region-based convolutional neural networks (R-CNNs), and single shot detection (SSD) have been used. Results are thoroughly analyzed and compared with each other in terms of precision, recall, and F1-score. It is found that faster R-CNNs outperform the other algorithms with an F1-score and mean Average Precision (mAP) values of 0.78 and 0.87 respectively.