Study on the Evaluation Index of Vehicle Tracking on Two-lane Road by Using Fractal Image Analysis

Traffic accidents at intersections account for almost half in Japan. If dangerous driving will be detected by tracking the vehicle nearby the intersection, it may be possible to decrease traffic accidents. It seems to be easy for drone to detect vehicles relatively at the right above intersection, but flight over the road is prohibited by the aviation law. Therefore, we have reported a method for detecting vehicles by performing fractal analysis on images taken from a safe location on the side of the road. However, our method detects and tracks vehicles by determining whether or not a vehicle exists within multiple continuous detection areas assumed in the image, so the existing vehicle tracking evaluation index IoU (Intersection over Union) is difficult to apply. In this study, we will consider the new evaluation index for vehicle tracking on Two-lane road by investigating the relationship between the ratio of vehicles included in the detection area and the image feature distance.