Research on Image Geo-Location in Parking Garage Based on Improved YOLO11

The performance of Global Navigation Satellite Systems (GNSS) is limited in weak-signal environments such as urban canyons and underground parking garages. Visual positioning, as an effective alternative solution, faces challenges due to the inefficiencies or inaccuracies of traditional methods (e.g., SIFT, ORB) and the computational redundancy of standard deep learning models (e.g., CNN) in specific simple scenarios. This paper proposes a real-time image-based positioning method tailored for underground parking garage environments. This method is based on the improved YOLO11 model. First, by replacing computationally intensive CBS modules with lightweight GSConv modules, the computational load and memory usage are reduced, effectively enhancing the model's operational efficiency on resource-constrained devices. Second, the improved model is integrated with Optical Character Recognition (OCR) technology to identify and localize parking space numbers as key visual features, providing a critical basis for positioning. Finally, by constructing a parking garage image database with precise location labels and a road network topographic map, useruploaded images are processed. Leveraging the enhanced YOLO11 model and OCR technology to recognize parking space numbers, the method achieves high-precision real-time positioning. Experimental results demonstrate that the improved YOLO11 model outperforms the original YOLO11, with significant optimizations in inference time (approximately 1.25 % reduction), memory usage (approximately 34.08 % reduction), and model size (approximately 12 % reduction). The proposed positioning method achieves high accuracy, real-time performance, and robustness, effectively addressing positioning challenges in specific weaksignal environments.