Using Optical Flow and Spectral Clustering for Behavior Recognition and Detection of Anomalous Behaviors

In this paper we propose an efficient method forbehavior recognition and identification of anomalous behavior invideo surveillance data. This approach consists of two phases oftraining and testing. In the training phase, first, we usebackground subtraction method to extract the moving pixels.Then optical flow vectors are extracted for moving pixels. Wepropose behavior features of each pixel as the average all opticalflow vectors in the pixel over several frames in video data. Next,we use spectral clustering to classify behaviors wherein pixelsthat have similar behavior features are clustered together. Thenwe obtain a behavior model for each cluster using the normaldistribution of the samples. Once the behavior models areobtained, in the testing phase, we use these models to detectanomalous behavior in a test video of the same scene.Experimental results on video surveillance sequences show theeffectiveness and speed of proposed method

Using Optical Flow and Spectral Clustering for Behavior Recognition and Detection of Anomalous Behaviors Keywords:

behavior modeling , optical flow , Gaussian distribution , anomaly detection , spectral clustering