Towards High Accuracy Parking Slot Detection for Automated Valet Parking System

Highly accurate parking slot detection methods are crucial for Automated Valet Parking (AVP) systems, to meet their demanding safety and functional requirements. While previous efforts have mostly focused on the algorithms’ capabilities to detect different types of slots under varying conditions, i.e. the detection rate, their accuracy has received little attention at this time. This paper highlights the importance of trustworthy slot detection methods, which address both the detection rate and the detection accuracy. To achieve this goal, an accurate slot detection method and a reliable ground-truth slot measurement method have been proposed in this paper. First, based on a 2D laser range finder, datapoints of obstacle vehicles on both sides of a slot have been collected and preprocessed. Second, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm has been improved to efficiently cluster these unevenly-distributed datapoints. After that, the Random Sample Consensus (RANSAC) algorithm has been improved to accurately fit the vehicles’ longitudinal contours. Finally, the candidate slot has been constructed and checked for its rationality. The final slot detection results have been defined in a way that contains both the slot size information and the slot relative position to the ego vehicle, which increases the requirement for detection accuracy. The performance of the proposed slot detection method has been verified on a test vehicle, and the experimental results show that the maximum errors in the detected slots under different conditions are 11.86 cm (position) and 3.35 deg (orientation).