- 25 views
- 133 downloads
Weapon classification using bounding box regression algorithms
-
- Author(s) / Creator(s)
-
Crime rates are increasing at a very high rate globally. The use of weapons in schools, airports, and streets is gaining popularity. To prevent this we have a surveillance system that is monitored by security officers and requires constant observation. This is a tiresome job and involves human errors regardless of the significance of the problem. This research studies different machine learning algorithms for weapon detection and classification from images and videos. We have classified the data into four segments: (1) Knife (2) Pistol (3) Rifle (4) Grenade. We have analyzed supervised learning models consisting of around 5000 images with manual labeling of each image. Due to the lack of images for every class, the images are selected from the internet manually. This study involves the analysis of deep learning algorithms such as VGG16, VGG19, ResNet50, DenseNet201, and MobileNetV3Large to predict the potential threat from the given input and identify the particular weapon with a bounding box. The training of the model involves pre-processing of images and extraction of various features of the image. Extracted features are then passed through convolutional layers to predict the output. Different performance metrics are used to compare the performance of the
deep learning detectors. The result of this study will help the government to establish an automated and reliable surveillance system. -
- Date created
- 2022
-
- Type of Item
- Research Material