Computer vision framework for visual sharp object detection using deep learning model

Abstract

Deep learning models are widely used for visual image feature extraction and classification. Troublemakers in human society may handle sharp objects like knifes, blades to perform crimes like burglary in public places. To monitor such activities, visual sharp object detection software needs to be integrated with camera based security and surveillance systems. To implement this application, our paper discusses about computer vision framework for sharp object detection using CNN model. Initially, object detection model was built using different CNN architectures namely AlexNet, ZFNet and VGG13. In order to improve the training and testing accuracy of the above models, a new CNN model was proposed with modified VGG architecture. The proposed CNN model has limited number of convolution layers with minimum weight parameters. Thus this model improves computation efficiency when executed on Intel CPUs and delivers better accuracy in training and testing when compared with other CNN architectures. Around 98% training and 92.2% testing accuracy was obtained for this model.