Abstract:
Weapon detection is a significant and intense issue in terms of public security and safety in general, and it’s undeniably a challenging and demanding
task. It’s even more challenging when you need to perform it automatically
or using AI models. Many object detection models have been developed that
accurately recognize items, but when it comes to weapons, it can be difficult to distinguish between weapons of varied sizes and shapes, as well as
the various colors of the background. Currently, many deep learning based
algorithms for real-time object detection are being developed. This research,
compared the two variants of the YOLO model in terms of weapons detection.
We generate a weapons dataset for training purposes, with photos sourced
from Google Images and a variety of other assets. We manually annotate the
photographs one by one in various formats, as YOLO requires a text-based
annotation file and some other models require an XML-based annotation file.
We used a large data set of weaponry to train both versions, and then analyse their results for comparison. We have elaborated in the research that
YOLOV4 performs better than the YOLOV3 for speed and accuracy. These
variants are also compared in terms of precision. Now these models are publicly available for better understanding. link:https://cutt.ly/4nVc1p2.
Index Terms-Weapons detection, computer vision, deep learning, object detection
