Classification and Detection of Abnormal Human Activity in Video Surveillance Based on SPATIO-Temporal Features

In this paper, we consider a human activity recognition approach that only requires single video example per activity. We introduce silhouette segmentation for extracting the foreground objects from background video. We used to represent a single human activity video by using bounding box representations, from that we can calculate height, width, aspect ratio, blob area, normalized bounding box specifying the sub-event’s spatial location, averaging the coordinates of all bounding boxes and Centroid, where human motion and body configuration is observed and tracked. We present an approach for measuring similarity between visual entities (images or videos) based on matching internal self-similarities .This intrinsic information is used together with statistical features and shape based approaches to recognize and classify Human activities. Spatio temporal features can be used for feature extraction of particular activity .Further; we had used K-NN, RVM, SVM classifiers in order to recognize the abnormal activity form human activity by using active learning algorithm and also measure the performance between them.