Detect Lane Line Based on Bi-directional Feature Pyramid Network
Jingqiao Tang · 2022 International Conference on Machine Learning and Intelligent Systems Engineering (MLISE) · 2022
The efficient detection and accurate identification of traffic signs is an important guarantee for the safe driving of intelligent vehicles, which has attracted the attention of many researchers. For the realistic traffic signs with small size and large scale changes, the performance of existing detection methods can not meet the application requirement of traffic signs detection. In this paper, based on the Yolov5 method, this paper propose a traffic signs detection algorithm based on the Yolov5 and weighted bi-directional feature pyramid network (BI-FPN). The BI-FPN helps combine the high semantic information at the top layer with high resolution information at the bottom layer to output a semantically rich feature embedding with clear localization information. Qualitative and quantitative results verify the effectiveness of this work, which can achieve effective detection of traffic signs in complex scenes. yolov5 is a one stage object detection tool with good detection accuracy for real-time detection tasks, however, it difficult to meet the requirements of multi-scale and low FLOPs in the process of traffic sign recognition.