Real-time vision system for nighttime vehicle detection

In recent years, the number of cars in the world has been increasing, resulting in a rise in the importance of video-based traffic flow monitoring and counting technology. However, compared with the computerized vision-based traffic flow monitoring and counting technologies used during daytime, those used during nighttime are less developed. In view of this, we have proposed a multi-feature technology, which can be used to monitor the headlights of cars during nighttime. First, the light source distribution in each captured image is analyzed and the length-width ratios of the headlights are distinguished. If two headlights both match the standards, a comparison between cars is carried out in order to avoid counting the same car more than once. The space between and colors of the two headlights are then used to judge whether these two lights belong to the same car using a similarity analysis algorithm. Experimental evaluations have shown that the proposed technology in this study can be stably used in both the monitoring of traffic flow and the real-time management of cars during nighttime.