Real time adaptive background estimation and road segmentation for vehicle classification

Our goal is detecting vehicle, lane of moving and classify in three types heavy vehicle such as trunk - semi heavy such as mini bus and van and light vehicle such as cab in order to recognize offender that is in illicit area. In this paper, we will propose an algorithm for moving vehicle detection different from previous works, detecting and classification carried out at several frame, this algorithm first segments best near field area of road according to camera installed point by Hough transform and k-mean line clustering. Then, this area is divided into many small non overlapped blocks and saved as a codebook and then extracted background of this area by adaptive background estimation. Vehicles is detected in two step, first kernel detection, second detecting sets that has the most correlation to this kernel, we have background subtracting and morphological binary image segmentation for vehicle detection by define of sets of ones connected together in binary image and identifying clustered sets that depend on single vehicle Occlusion detection and elimination and define type of it by energy of clustered sets on day and distance backlights and distance between backlights and headlights at night and a neural network classification. Proposed algorithm has a satisfying performance under various conditions, which can robustly and effectively eliminate the influence of casting shadows, headlights, or bad illumination and occlusion.