Dual model-based traffic light and sign detection using prior information

Traffic light and traffic sign detection are important in the field of self-driving. They can guide vehicles to drive safely on the road. It is difficult for existing algorithms of object detection to detect targets simultaneously and achieve high accuracy. In this paper, a dual-model framework is proposed to detect traffic light and signs for a self-driving vehicle based on prior information. This framework can switch the detection model according to the prior information. The color information of the traffic sign is used to extract the ROI and improve the detection efficiency. The work of this paper also includes collecting and annotating a large amount of image data to apply the model trained on the proposed framework to self-driving. The proposed framework is verified on a real road test of a self-driving vehicle. © 2020 Totem Publisher, Inc.