Abnormal Activities Detection from Video Surveillance Using the Auction Optimization Based LPBoost Convolution Neural Network

Authors

  • Giriprasad S, Mohan S , Vani V

Abstract

Video surveillance is the process of monitoring the various activities and behavior of the anomalies in the crowd
sequences which is used to monitoring and protecting the environment. The video surveillance is used in various
applications such as crowd monitoring, education institutional and hospital environment so, on. In the captured
video, anomaly activities has been difficult to identified because of the high density of the crowd which is placed an
major challenge while monitoring the activities in the crowd. So, in this paper, the difficulty is handled by using the
Auction optimization based LPboost Convolution Neural Network. The anomaly detection and classification process
are performed by background subtraction, object detection, feature extraction and classification phases. From the
video frames, the background has been eliminated by fuzzy based multilayer subtraction approach and the objects
identified using the Principal Bow based Region Descriptor. Then the extracted patches are grouped by Supervision
based Similar Patch Clustering method in which patches are grouped based on their shape and direction. Finally, the
anomalies are classified by using the Auction optimization based LPboost Convolution Neural Network which
enhances the performance of the system using the optimization and boosting method. The efficiency of the proposed
system analyzed with the help of the Saface crowd video surveillance data sets and the comparison are made
between the existing systems such as Neural Networks, MLP, KNN, Background Mixture Model, Gaussian Mixture
Model and so on which shows that the proposed system has a higher accuracy level.

Published

2020-03-31

Issue

Section

Articles