基于双色像素的Hough直线检测算法

Hough straight line detection algorithm based on two-colored pixel

针对Hough变换检测计算量巨大,而且检测结果含有大量虚假直线的问题,利用双色像素(TCP)与矢量内积过滤作为Hough变换输入的边缘点。将图像划分为若干子图,利用TCP算法得到每个子图中潜在的直线段;计算Canny边缘点的梯度方向;再计算边缘点梯度与其所在子图直线段方向矢量之间的内积,作为判定边缘点是否位于直线上的依据。为了提高算法的效率,可用CUDA并行分层算法加速TCP的计算。由于过滤了大量的边缘点,使得Hough变换的计算量大大减少,同时提高了Hough变换的直线检测准确度。实验表明,该算法可以高效、准确地检测出图像中的主要直线。

In view of the large amount of calculation of the Hough transform, and the problem of lots of false straight lines contained in the detection result, a novel algorithm was proposed which used TCP （ Two-Colored Pixel） with vector inner product to filter the input edge-points of the Hough transform. Firstly, an image was divided into several subimages, TCP algorithms was used to get the potential straight segments in each subimage; secondly, the gradient direction of each edge- point detected by Canny operator was determined, and then the inner product of the gradient of each edge-point and its subgraph TCP-line＇s direction vector was calculated; the value of inner product for an edge-point was finally served as its determinant for whether it lied on a straight line. In order to improve the efficiency of the algorithm, CUDA parallel layered algorithm was exploited to accelerate the calculation of TCP. Because a large number of edge-points were filtered, the amount of calculation of the Hough transform was greatly reduced, and as a result the accuracy of straight line detection was also to be improved. Experiments show that the algorithm can efficiently and accurately detect the main straight lines of an image.