Clustering and forecasting of Region of Interest by dividing screen into meshes in video frames

Image processing and security surveillance system has more and more widely used in recent society such as bank surveillance and pedestrian tracking. And there is a common scene that each frame in the video is constructed by a set of pixels. However, the size of pixels are fixed. And the detection of Region of Interest (Rol) is always been regarded as the most significant in tracking system. Based on this, we proposed a concept that divide the whole frame into a set of granule units and detecting Rol within certain units instead of the whole frame during the processing. The size of which can be modified to fit the different situations. The authors applied one of the algorithm which called 'Density-Based Spatial Clustering of Application with Noise' (DBSCAN), and combined it with foreground detecting algorithm 'Kernel Density Estimation'(KDE) and tracking algorithm 'Kalmen Filter' to testify this concept, and got the ideal result in the experiment.