For intelligent vehicles, if the sensing device can accurately and quickly detect the protruding speed bumps on the road ahead of the car in a harsh light environment, it can provide important pre-information for the intelligent vehicle control system and ultimately improve the vehicle’s overall performance. The speed bump data sets under different environments and perspectives were collected and produced to address the problem of low detection accuracy and missed detection in detecting speed bumps on the road on low-light and rainy days. Based on the YOLOv8n model, the SE attention mechanism was first added to the backbone network to improve the feature expression ability. The EMA attention mechanism is added to the neck to effectively highlight the contour features of the detection target under a complex background. Finally, a small target detection layer is added to detect long-distance and broken speed bumps more accurately. Offline simulation experiments show that the improved model improves the detection accuracy mAP@0.5 by 8.9% compared with the original model when the calculation parameters are almost the same. The enhanced model is deployed to the experimental vehicle, and the speed bump in front of the car can be effectively detected by verifying the model. The results show that the proposed improved method effectively improves the speed bump detection accuracy under low-light conditions and can accurately provide the pre-information for speed bump obstacle detection, which has good scientific research significance and application value.
Li et al. (Mon,) studied this question.