Intelligent video surveillance using enhanced deep belief based multilayered convolution neural network classification techniques

Abstract

Abstract Video surveillance has undergone numerous changes in the past few years and several pieces of research have been carried out in this field. Object tracking is the significant task in such systems, and hence it is essential to review the standard approaches dealing with object detection, classification and tracking. This work proposes a novel classification technique for a detected object of a moving scene from a video dataset. Initially the dataset has been processed and prepared for data training based on neural networks. The data has been classified using the proposed enhanced deep belief based multilayered convolutional neural network (EBMCNN). The major focus is on deep learning applications involved in estimating the count, total persons involved and the activities in a crowd where all criteria are taken into consideration, thereby achieving security through video analysis. Identifying theft and the detection of violence are some security measures where video is converted to frames which are then processed to analyse the individuals along with their activities. The classification is performed through comparative analysis of a real‐time dataset. Experimental results show the accuracy 97%, precision 93.8%, recall 87.7% and F‐1 score 87.5%.