A Multitask Method for Parking Slot and Obstacle Detection using YOLO

Scene awareness is fundamental to automatic parking systems, where parking slot detection and obstacle detection provide accurate environmental perception. Existing studies primarily address these tasks separately, whilst unified models remain insufficiently explored. A multi-task framework is proposed based on a one-stage detection paradigm to jointly address parking slot detection and obstacle detection, as these tasks exhibit heterogeneous feature requirements. The YOLOv11 architecture is extended with lightweight modules and attention-based feature modulation, together with a dedicated detection head, to improve task-specific feature representation. Experimental results on the dataset demonstrate a precision of 95.05% and a recall of 96.26% for parking slot detection, and a precision of 93.97%, a recall of 96.38%, and a mAP@50 of 97.35% for obstacle detection. Real-vehicle experiments further validate the robustness of the proposed method in diverse road and parking scenarios.