Target distance measurement method in automotive assisted driving based on improved Yolov8

Begin the abstract two lines below author names and addresses. The abstract summarizes key findings in the paper. It is a In the research of autonomous driving, detecting target objects and measuring the distance between them are hot topics in assisting drivers in predicting risks. Due to the fast speed of vehicles on the road and the small and difficult to detect characteristics of pedestrians and other targets, it is difficult to achieve real-time detection and accurate distance measurement using the original object detection model combined with monocular distance measurement methods. In response to this issue, this paper proposes an improved Yolov8 object detection method based on binocular ranging. The model introduces the cot3 module and improves the loss function. The target ranging part uses the SGBM matching algorithm with improved performance balance. The experimental results showed that the improved Yolov8 model achieved an accuracy of 0.858, which is 5% higher than the Yolov8 model. Within the effective detection range of the binocular camera, the maximum absolute error of target ranging is 0.39 m, and the maximum relative error is 2.11%. The system realizes target detection and distance measurement, and can achieve safe real-time distance measurement and alarm function for autonomous driving.