Real Time Analysis of Video Behaviour Profiling for Detection of Abnormal Activity in Frames

The field of analyzing Video Behaviour profiling is very keen and distinct part in the area of Computer Vision and Video signal processing. The two main attributes of this field are Human Action Recognition (HAR) and Abnormal Activity Detection (AAD). Various researchers proposed several algorithms for HAR and AAD, since the analysis of video behaviour profile has numerous applications. This paper aims to detect the frames of abnormal activity in the surveillance video using combination of Optical flow vectors with morphological operation on the foreground pixel data. Prior to extraction of foreground pixel data, contrast adjustment applied to the frames for robust identification of motion based pixels only. This novel concept further uses the threshold of pixel intensity values in terms of motion and its continuity over a period of time to identify abnormal activity in consecutive frames. The proposed algorithm is computationally efficient in detecting abnormal activity frames, which is an attractive feature for real time applications.