A new approach for circular traffic sign tracking from image sequences

抄録

This paper describes a new approach for tracking circular traffic signs from image sequences. As a part of traffic sign recognition system, the traffic sign tracking will improve the performance of recognition. By tracking the signs, search space is reduced and misdetection caused by temporal occlusion could be avoided. We propose a new blob tracking called two-layer blob tracking to track the signs from frame to the next. This tracking does not require an accurate model, which is essential in the Kalman-Filter tracking. A red color thresholding followed by a blue color thresholding are used to produce the two-layer blobs. To track the blobs between two successive frames, a blob matching using ring-partitioned matching is employed. It could match the rotated, resized and occluded blobs effectively. The proposed tracking method could track the traffic signs from a moving camera, without have to consider some restrictions on the speed and movement of vehicle, and camera installation.