A deep learning behavior analysis model for efficient video surveillance using multi pose features

Abstract

Organizational security is addressed via a variety of ways, with video surveillance listed as a crucial element. Many firms use video surveillance to monitor their environments and processes, as well as to enforce security. Several designs have previously been presented, which use action templates, shapes, and other aspects to identify the activity. However, the proposed CCMMBAM model need to perform better in the classification of activity. OKUTAMA-Action collection contains images of humans in several poses. The approach uses continuous categorization to do video surveillance and increase security. Input photos are segmented using Modified threshold-centric K-means clustering. In each session, the process keeps distinct activity templates for different positions. The features were then optimized using a quadratic discriminate analysis, followed by training a convolutional neural network to accomplish the action classification. Similarly, the approach retrieves information such as edge mobility, gray parts, and a binary pattern. As a result, the neurons are programmed to calculate Behavioral Edge Support (BES), Behavioral Pattern Support (BPS), and Behavioral Gray Covariance Support (BGCS) for various classes of actions based on the activity templates provided by the system. The approach uses continuous categorization to do video surveillance and increase security. Finally, the proposed CCMMBAM model resulted in improved categorization and object tracking accuracy.