A real-time and color-based computer vision for traffic monitoring system

This paper presents a vision-based traffic monitoring system via analyzing color image sequences of traffic scenes recorded by a stationary camera mounted on a tall building or pedestrian crossing bridge near a traffic light. The system algorithms consist of building initial background, segmenting foreground, shadow separation, vehicle location, vehicle tracking, and background updating. The YCrCb color space is adopted in our algorithm. It provides us an accurate and robust device for foreground and shadow detection. Our system can accurately locate and track vehicles from image sequences. It is fast enough to operate in real time, insensitive to lighting and weather conditions, and only requires a minimal amount of initialization. Experimental results from town scenes are provided which demonstrate the effectiveness of our system.