Multimodal Road Traffic Detection Algorithm based on Improved YOLOv8

Abstract

In road traffic detection, traditional unimodal object detection methods exhibit certain limitations in adapting to environmental variations. Moreover, in complex road conditions, mutual occlusion between targets and their confusion with the background pose significant challenges in feature extraction for multi-scale objects and the detection of densely distributed small targets. To address these challenges, this paper proposes a multi-modal object detection algorithm, CD-MMNet, based on YOLOv8. Firstly, the backbone network adopts a dual-branch structure to perform intermediate fusion of features from two modalities—visible light and infrared images—thereby leveraging their complementary characteristics to dynamically select optimal feature extraction in a targeted manner. Secondly, the CBAM attention mechanism is introduced to dynamically adjust the importance of each channel and spatial position in the feature maps, enhancing key regional features while suppressing background noise, thus improving the model's feature extraction capability. Finally, the DBB module is incorporated, utilising a diversified branch network to enhance the model's adaptability to feature maps of varying scales. Experimental results demonstrate that the proposed algorithm outperforms the original YOLOv8 and other mainstream algorithms on the M3FD dataset, achieving a 4.0% improvement in mAP@0.5~0.95 compared to the baseline YOLOv8. This effectively enhances object detection performance in challenging environments such as adverse weather conditions and traffic congestion.