An Abnormal Activity Detection Using Harris corner, Optical Flow and Dynamic Thresholding

Abstract

In this paper, new method has been introduced for identifying abnormal activity from videos. This method is totally based on motion which can be calculated by applying optical flow method. In this method key points are also detected for calculating method. Key points are detected by Harris corners. For identifying abnormal activity dynamic threshold is used. For calculating dynamic threshold energy has been used. This method doesn’t used any prior knowledge or training data. It also do not required static threshold for identifying abnormal activity. This method can detect abnormality from UMN, BEHAVE and UCF-Crime dataset. The activity identified robustly. It also doesn’t require much computation.