YOLOv10n-BC:a novel real-time object detection model for Driver distracted driving detection

Abstract

Driver distraction is a primary cause of traffic accidents, and the real-time and effective detection of such behaviors can significantly reduce traffic-related injuries and fatalities. In this paper, we enhance the lightweight YOLOv10n model by integrating the BiFPN structure to bolster its multi-scale feature extraction capabilities. Additionally, we design a CASSA module that combines channel attention, spatial attention, and channel shuffle to strengthen the model's ability to capture long-range dependencies. The model was tested on the CBTDDD dataset, established in this study, which includes data on driver distraction across multiple scenarios involving sedans, passenger buses, and trucks. Compared to the original YOLOv10n model, the proposed model demonstrates a 2.0% improvement in mAP@0.5 and achieves an FPS of 115.3 f/s. These results indicate that the YOLOv10n-BC model developed in this paper is capable of performing real-time and efficient monitoring of driver distraction.