Improved Brake Light State Detection Algorithm of YOLOv8
XU Jian
LIU Yu
Abstract：Smart car light self-closing loop requires rich perception results of the vehicle's surrounding environment,while meeting real-time,high accuracy,and low omission(error)detection rates.Based on this,a lightweight detection algorithm model,YOLOv8-Brake Light(YOLOv8-BL),based on YOLOv8n,is proposed.This method enhances the model's ability to extract complex environmental target features adaptively by adding the Squeeze-and-Excitation(SE)module.At the same time,lightweight convolution is introduced in the Conv module,replacing ordinary convolution with the GSConv module,effectively reducing the model's computational load.On the basis of the GSConv module,a single aggregation module,VoV-GSCSPC,is designed to optimize the model structure and achieve higher model performance.Experimental results show that the YOLOv8-BL model has a higher mAP@0.5 index on the private front-view dataset than the YOLOv8n model,an improvement of 3.9%,and the computational load is reduced to 7.6 GFLOPs.The proposed YOLOv8-BL model can effectively complete the smart car lights perception task at the edge end,and has good practical application value.
Keywords：nighttime perceptioncar light self-closing loopchannel attentionYOLOv8
Publication Date：2025-10-25
Online Publishing Date：2025-11-10(First online date of this platform, not the publication date of the document)
Pages：6( 71-75,79 )