基于图像梯度匹配的自动泊车系统车位识别方法

Parking slot recognition of automatic parking system based on image gradient matching

针对自动泊车系统利用视觉方法识别车位线的需求,提出了一种基于图像梯度和高斯加权余弦距离的车位线识别算法.首先对采集的车位图像进行去畸变和逆透视变换生成俯视图,并进行灰度化和图像滤波处理;然后利用Sobel算子提取预处理后的图像的梯度,选取车位线的车位角位置,建立图像梯度模板,并利用双线性插值法建立旋转模板以识别不同角度的车位;最后以高斯加权余弦距离作为相似度量,识别图像中车位线的位置,设计了算法的终止条件以提高算法运行速度.结果表明:该算法实现了对不同工况下多种车位的识别,具有识别速度快、不易受光照路面裂纹等因素影响的优点.

According to the requirement of automatic parking system for identifying the parking slot by visual method,a parking slot recognition algorithm was proposed based on image gradient and Gaussian weighted cosine distance.The acquired parking space image was de-distorted,and the inverse-perspective transformation was performed to generate a top view.The gray processing and the image filtering processes were performed.The gradient of the preprocessed image was extracted by the Sobel operator,and the parking space position of parking slot was selected.An image gradient template was created,and a rotation template was created by bilinear interpolation to detect parking spaces at different angles.The Gaussian weighted cosine distance was used as the similarity measure to detect the position of parking slot in the image.The termination condition of the algorithm was designed to improve the speed of the algorithm.The experimental results show that the multiple parking spaces can be recognized under different working conditions by the proposed algorithm with fast recognition speed and not easy affecting by cracks on the road surface.