Light Model based on End-to-End for Steering Angle Detection

Abstract

This paper performs lane line angle detection in autonomous driving scenarios based on an end-to-end learning mechanism. Lane line angle detection is a vital technology research development direction in Autonomous Vehicles. However, since most lane line targets in remote sensing images have sparse features, it is still challenging to achieve accurate lane line angle detection in traffic status images in front of vehicles. A lane line angle detection algorithm based on the improved C3 module YOLOV5n algorithm is proposed, which mainly includes: a self-made lane curvature dataset; improvement of the loss function; improvement of the C3 module to improve the detection accuracy of the network. Experiments are conducted using the traffic status images in front of vehicles in the lane curvature dataset, and the results show that the algorithm achieves better detection results in lane curvature detection.