Virtual Line Group Based Video Vehicle Detection Algorithm Utilizing Both Luminance and Chrominance

Abstract

The virtual line based video vehicle detection algorithm is used extensively in intelligent transportation system (ITS) because of its high real-time performance, but it only utilizes luminance values of pixels on only one virtual detection line to detect moving vehicles, as a result, its false reject rate (FRR) and false accept rate (FAR) are high. In order to solve this problem, a virtual line group based algorithm using both luminance and chrominance information and the spatiotemporal information of the virtual line group is proposed as an improvement for it. The virtual line group includes four adjacent virtual detection lines. The improved algorithm also bases on the method of background subtraction to detect vehicles on each detection line as the original algorithm, that is, subtracting the background from the current frame, then comparing the differences with the threshold to judge whether there is a vehicle. But the improved algorithm introduces two-level detection to do this, that is, it first does the first level detection utilizing luminance information as the original algorithm does, if it decides there is no vehicle it will modify the luminance threshold according to the pixels' colors acquired by using the chrominance information to do the second level detection. After that it optimizes the result using the spatiotemporal information of the virtual line group. The experiment results demonstrated that the improved algorithm could effectively reduce the FRR and FAR, and it was more accurate and robust.