Robust vehicle detector using spatio-temporal MRF model
Daisuke Ushirogohchi, Yoshihiro Sakamoto, Koichiro Kajitani, Takeshi Naito, Shunsuke Kamijo · 19th ITS World CongressERTICO - ITS EuropeEuropean CommissionITS AmericaITS Asia-Pacific · 2012
Nowadays, cameras set on highways and local streets are broadly used in traffic measurements and monitoring systems. Traffic measurement plays an important role in ITS (intelligent transportation systems), which can provide congestion and travel time information for drivers. However, traditional methods suffered the problem of occlusion and illumination effects. In this paper, a novel vehicle detector is introduced and the experiment results show that the authors vehicle detector performs robustly under various environmental conditions. In the authors vehicle detector, based on the high performance of vehicle tracking, the traffic flow of each lane and velocity of each vehicle can be calculated precisely. As for the vehicle tracking, the ST-MRF model is used which is very effective even in circumstances of occlusion and illumination change. Considering low illumination conditions at night, a tail lamp extraction algorithm is developed for vehicle detection. Experiments under various conditions show that our vehicle detector achieved a detection rate of 97.9% in total.