Abstract:

An algorithm was proposed herein for an automatic parking system, which used vision for recognition of parking slot marking. Pyramid hierarchical search strategy was adopted. First, image binarization was carried out by applying K means clustering to the intensity histogram and skeleton of parking slot marking was extracted. Parking space corner candidate point was determined in high-level pyramid image using Hough transform to detect skeleton and skeleton clustered using unsupervised cluster based on density. Then, region of interest was seleted in lowest-level pyramid image and skeleton was extracted using improved distance transform-based skeleton extraction algorithm. Skeleton of parking space corner was matched using genetic algorithm and target parking space was determined according to characteristics of  actual parking space. Lastly, the effectiveness and rapidity of the proposed algorithm was verified by collecting many parking space images in the outdoor environment. Experimental results demonstrate that computation is small and parking recognition accuracy is as 98% using the proposed algorithm.

Key words: automatic parking;vision;parking , slot , marking , recognition;pyramid delamination;image , matching

摘要：

提出了一种自动泊车系统中采用视觉方法通过识别车位线来确定泊车位的算法。采用金字塔分层搜索策略，首先，在灰度直方图上应用K均值聚类法对图像进行二值化，提取车位线骨架，采用Hough变换检测骨架，并利用基于密度的无参数聚类方法对骨架线聚类，在金字塔高层图像上确定车位角点候选点;然后，在金字塔最底层图像上选择感兴趣区域，采用改进的基于距离变换的骨架提取算法提取骨架，使用遗传算法对车位角点骨架进行精确匹配，根据实际车位角点的分布特征确定目标车位;最后，在室外不同环境下采集多张车位图片进行算法的有效性和快速性验证实验。实验结果表明,采用基于视觉的车位线识别算法进行车位检测能较大地提高检测的效率和识别正确率。

关键词: 自动泊车, 视觉, 车位线识别, 金字塔分层, 图像匹配