Detection of dropped non protruding objects in video surveillance using clustered data stream

As more and more surveillance cameras are deployed in a facility or area the demand for automatic detection of suspicious objects is increasing. Most of the work in recent literature concentrated on protruding object detection in video sequences. This paper proposes a novel approach to detect protruding as well as non protruding objects in sequences of walking pedestrians based on texture of the foreground objects. Initially static background is modeled with the help of mixture of Gaussian algorithm and the foreground objects are segmented. Later object is detected frame by frame which is followed by the calculation of statistical parameters such as mean and standard deviation, in every blob, to form data streams. These parameters are clustered online using k-means methodology over data streams, in order to find the outliers (dropped objects). Here k is based on the number of objects present in the video. Finally we have implemented on a standard data set from the website Video Surveillance Online Repository [15] and also our own dataset. The experimental results show that our system performs reasonable well and can accurately detect dropped objects in video data streams.