Study of Novel Neural - Fuzzy Logic for Abnormal Behavior Detection

An abnormal behavior detection algorithm for surveillance is used to identify the targets as being in a normal or abnormal movement. This work falls within the framework of the video surveillance research axis. The job is to analyze video streams coming from a network of surveillance cameras, deployed in an area of interest in order to detect abnormal behavior. A study of novel model is designed here for this purpose. The uniqueness of this novel algorithm is the use of foreground detection with Gaussian mixture (FGMM) model before passing the video frames to optical flow model using Lucas-Kanade approach. Information of horizontal and vertical displacements and directions associated with each pixel for object of interest is extracted. These features are then fed to feedforward neural network for classification, to decrease the rate of false positive the output of neural classification further classified by use of fuzzy logic. The study is being planned to conduct on the real time videos and some synthesized videos. The detection of these behaviors will increase the speed of response of the security services in order to perform accurate analysis and detection of events in real time.