Research on object detection in autonomous driving road scene based on improved YOLOv11 algorithm

With the growing adoption of autonomous driving technology in the automotive industry, vision-based object detection has emerged as a critical component of autonomous systems. This study enhances the YOLOv11 algorithm for road scene object detection using the Udacity Self-Driving Car dataset processed by Roboflow. The proposed improvements include: integration of an additional detection layer and specialized detection head for small targets, replacement of the original C3k2 module with an RCS-OSA module to enhance feature extraction capability, and implementation of the PAA attention mechanism to optimize feature fusion accuracy. Experimental results demonstrate that the modified YOLOv11 achieves significant performance improvements, showing 4.6% increase in [email protected], 7.6% gain in mAP 50-95, and 6.4% enhancement in recall rate compared to the baseline. These modifications effectively address key challenges in detecting distant small targets and dense object clusters within complex driving environments.