색상 적응적인 차량 그림자 제거방법

Abstract

The use of video image processing in vehicle detection systems is a widespread trend in the field of Intelligent Transportation Systems (ITS) because of low costs, fast processing and convenience of installation and application. In this paper, we propose a new method for increasing the performance of ITS systems. This method removes the shadow area of vehicles and reduces the noise of video images. The proposed algorithm, which removes the shadow area of vehicles and reduces the image noise is conducted in HSV color model because of its characteristics. The algorithm to detect the shadow area of vehicles consists of background extraction and update, initial vehicle detection, color-adaptive shadow removal, outline removal, and vehicle object acquisition. Lee's method is utilized in two steps to detect the foreground and create the accurate background. In the post-processing, the algorithm fills holes on the vehicles and removes the outline of the vehicles. In our experiments, we categorized vehicle images into five classes, including white vehicles, black vehicles, vehicles on the straight roads, vehicles in intersections, and vehicles on the stop lines, to test how robust the proposed algorithm is exposed variation in chromaticity and brightness as well as variation in road shapes. To evaluate the performance of the proposed algorithm, we compared it with three existing methods, including NRGB, CHID, and CUCHIARA. F-measure is used to evaluate the performance of the algorithm when removing vehicle shadows. The measure is based on the Cohen’s DR (Detection Rate) and FAR (False Alarm Rate). The proposed algorithm achieved superior performance with the average DR of 93.1%, FAR of 33.1%, and F-measure of 77.0% when compared to existing methods.