基于环视系统的停车位检测与跟踪算法

Parking Slot Detection and Tracking Algorithm Based on Around-View System

为提供准确有效的停车位位置信息,提出一种基于车位导轨线跟踪和车位角特征的检测方法。首先,从直线段检测(LSD)算法检测的线段中提取车位导轨线,通过导轨线跟踪算法增强环境适应性,确定车位角的待检测感兴趣区域。其次,训练了单个基于方向梯度直方图(HOG)特征的支持向量机(SVM)分类器进行“T”型和“L”型车位角检测,使用改进的非极大值抑制算法优化检测结果。试验结果表明,与仅基于车位线特征或车位角特征的算法相比,所提出的算法漏检率更低,在多种天气条件下具有较好的鲁棒性和适应性。

To provide accurate and effective parking-slot position information,a detection method is proposed based on entrance-line tracking and marking-point feature.Firstly,to determine the region of interest of marking-point detection,the entrance-line of parking-slot is extracted from the line segments detected by the LSD algorithm,and the environment adaptability is enhanced through the entrance-line tracking algorithm.Secondly,a single classifier is trained to detect the“T-shaped”and“L-shaped”marking-point,using HOG feature and SVM algorithm.A modified Non-Maximum Suppression algorithm is applied to optimize the detection results.Experiment data show that the proposed algorithm has lower missing rate compared with methods only based on parking-line feature or marking-point feature,and the algorithm has robustness and universality to deal with various weather conditions.