Accurate vehicle detection and counting algorithm for traffic data collection

Number of vehicles on road is very important traffic data and is essential for transportation safety and management. In this paper, an approach for vehicle detection is presented. In this approach, virtual line based sensors which are just straight detection lines are first set on road lanes. Then two features, namely gradient and range feature, are proposed for vehicle detection. This is carried out by extracting and analyzing the two features on detection lines. Meanwhile, the solution for vehicle occlusion have also been proposed. Our proposed method has an outstanding advantage that it performs excellent in traffic jams as well as under various conditions, such as sunny, cloudy, and rainy days, or night time, or even tunnels with complex illumination. The high accuracy rate of our method is verified with the experiment results.