Detecting and localizing the anomalies in video surveillance using deep neuralnetwork with advanced feature descriptor

Abstract

The anomaly detection in the video surveillance is the need of the hour to detect the anomalous events or objects under the emergency conditions. The main aim of the research is to develop an automatic anomaly detection strategy for the detection of anomaly in surveillance videos. Initially, the input surveillance video is subjected to object detection using threshold method and the object tracking is performed using the Minimum output sum of squared error tracking algorithm (MOSSE). Once the objects are tracked, the statistical and advanced texture features are extracted using the statistical models and advanced texture descriptor. Then, the feature vector with statistical and textual features is fed as input to the deep convolutional neural network (DCNN) classifier that classifies the video as normal or abnormal. If any anomalies are detected in the surveillance video, the object localization is done to track the location of the abnormal object in the video. The comparative analysis of the methods reveals the better performance of the deep CNN classifier with a higher accuracy of 92.15%.