A method to improve the accuracy of vehicle identification by perspective transformation
Zhang Xing-lon · Electronic Design Engineering · 2015
Too many types of moving targets pose a great challenge to accurate statistics of vehicles based on video. In this paper, Gaussian background modeling and background subtraction are used to detect the moving objects. Then, perspective transformation is applied to the image. This results in that a unified threshold can be employed to identify the vehicles from other non-targets such as pedestrians, bicycles and motorcycles due to their different areas. The proposed algorithm has a low computational complexity. Experimental results show that the method can shield non-targets well and reduce the false detection rate of vehicles.