VGG-SSD Model for Weapon Detection using Image Processing

Video surveillance demands application of technology like deep learning, image processing for automated detection of deadly weapons in the hands of anti-social elements. This work depicts the application of very deep convolutional neural network (CNN), Visual Geometry Group (VGG)-16, and popular object detection technique, Single Shot MultiBox Detector (SSD) for detecting a weapon in automated surveillance application. The model is trained and tested on benchmark data on normal, gun, and knife objects. Performance parameters are, Mean Average Precision (MAP), Recall, Intersection over Union (IoU), and classification loss. A MAP of 87%, a recall of 86.6%, and calculation loss of 0.07 and total loss of 0.35 are achieved and the model proves to be efficient.