Traffic counting errors due to occlusion in video image vehicle detection systems

Video Image Vehicle Detection Systems (VIVDS) are becoming primary tools of detecting traffic at intersections, interchanges and freeways due to their low costs, ability to monitor multiple lanes, variety of traffic data collection functions, and ability to easily reconfigure. However, the performance of VIVDS is often affected by camera location, lighting, weather and other factors. This paper will examine the effects of occlusion on traffic counts on freeway segments. Occlusion is a common phenomenon for VIVDS and the data errors are caused by many factors. There are two types of occlusion: longitudinal occlusion which causes missing detections, and horizontal occlusion which causes false counts. This paper also includes models that quantify missing and false counts. Finally, the models were validated using data collected on the I-80 freeway in Reno, Nevada. The test proved that the models are reliable and useful for real applications.