Two-stage vehicle leaving and entering detection for ground truth generation

The purpose of ground truth data is to provide an absolute reference to be compared with algorithms´ results. However, collecting ground truth data for real-life applications is an important but very disturbing task. This is particularly true for license plate recognition applications because of time-consuming manual. In this paper, we present a method of two-stage vehicle leaving and entering detection for ground truth generation in video sequences. The mechanism started from a general license plate recognition system to recognize the numbers. Then, we proposed a two-stage method to detect vehicle leaving and entering based on particle tracking. Moreover, the tracked regions would be used to reduce the searching areas of the next input frame. Once the vehicle is out of the camera field of view, the timestamp of the vehicle entering is detected by the vehicle entering detection procedure. Finally, the ground truth data of the passing vehicle is collected with the three attributes, i.e., the timestamps of entering and leaving and the identity numbers of the vehicle. In our experiments, all frames with the license plate heights ranging from 12 to 40 pixels were captured from an entrance at a University. The result shows that the proposed method can achieve low false alarms rate and miss rate.