A vehicle image tracking method robust to size reduction and partial color change

In recent years, vehicles have a driving support device which recognizes a forward vehicle for auto cruise or collision avoidance. Since these devices are related to human safety, they are required sufficient evaluation in their development process, but evaluation data preparing cost is too high because it needs to prepare manually on human recognition. Therefore, we propose an image tracking method useful to reduce human's annotation work. This method has robustness to size reduction and partial color change, which are apparent changes often occurring in a forward vehicle. Experiments showed that this method enables to generate accurate position annotations of many frames from the position annotation of 1 frame.