Traffic sign detection algorithm based on adaptive enhancement of low-illumination images

Résumé fourni par la source

To address the issue of traffic sign detection under low-light conditions, this paper proposes an Adaptive Image Enhancement-based Traffic Sign Detection Algorithm (AIE-YOLO). In low-light environments, road traffic sign images often suffer from issues such as blurred features and significant background noise, which degrade the performance of object detection. To tackle these challenges, this paper designs an image adaptive enhancement module based on Retinex theory. This module automatically adjusts the critical hyperparameters of the Retinex image enhancement algorithm through a parameter estimation network, enabling adaptive enhancement of low-light images according to lighting conditions. Additionally, to improve the detection capability for small objects, large kernel depthwise separable convolutions are introduced into the Backbone of YOLOv8s, and a lightweight upsampling operator, CARAFE is utilized in the Neck layer to mitigate the loss of low-level features during upsampling. Experimental results demonstrate that the proposed algorithm is highly effective for detecting traffic signs in low-light environments, with fast frame rates that meet the performance requirements of low-power embedded devices.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.