A framework of traffic lights detection, tracking and recognition based on motion models

Abstract

Detection of traffic lights is a basic technology for autonomous vehicle and driver assistant system. This paper presents a framework of detection, tracking, classification and online mapping using the images captured by a camera mounted on the vehicle and the position and attitude information from GPS/INS. The sequential results of detection, which is treated as observations with uncertainty, are associated with the targets in previous frame. The results of association are filtered and classified. In addition, the target position in the image is predicted based on a novel motion model and aided by a online mapping module that provides the model with 3D location information. The precise motion model significantly improves the performance of the association. The prediction algorithm based on our motion model is evaluated and compared with the other methods.