Research on a Detection Algorithm of Contaminant on Automotive Cameras Optical Surface Based on SEFaster-YOLOv8

Cameras are crucial sensors in intelligent driving systems. Due to the optical windows of these cameras generally being exposed, they are highly susceptible to contaminant from external dust, mud, and other contaminants. These contaminants can degrade the vehicle’s perception capabilities, posing safety risks. Therefore, research on the identification and automatic cleaning of optical window surface contamination for automotive cameras is essential. This paper constructs a dataset of contaminated images of automotive cameras using a method based on shooting and image fusion. By introducing the SE attention mechanism and replacing the YOLOv8 backbone network with FasterNet, this paper proposed the SEFaster-YOLOv8 model. Experimental results show that the SEFaster-YOLOv8 model reduces the parameter count by 37.6% compared to the original YOLOv8 model. The mAP@0.5 and mAP@0.5:0.95 reach 95.7% and 66.9%, respectively, representing improvements of 1.8% and 1.1% over the original YOLOv8 model. The fps reaches 110.11, a 22.33% increase compared to the original YOLOv8 model.