一种基于边缘图像的快速物体检测方法

Fast object detection method based on edge image

为提高物体检测速度,提出一种利用边缘快速检测物体的方法。首先,求取图像中每一个像素的边缘模值和方向,依据模值定位边缘位置,依据位置和方向进行边缘分组;然后,通过计算边缘组的相似度来对候选的包围盒进行投票;最后,统计多尺度滑动窗口上各候选包围盒的投票得分,依据包围盒得分对包围盒进行过滤,再依据包围盒重合度对包围盒进行合并,并依据合并的包围盒数量判断其是否为物体,最终得到图像中各物体的包围盒。实验结果表明,相对于目前最新的物体检测算法,该方法的运算效率较高,且检索率和精确度高。

For improving the speed of object detection, this paper proposed an object detection method by using edge. First, it computed the edge magnitude and orientation for every pixel in the image, located the edge position according to magnitude, and divided the edges into different groups according to edge position and orientation. Second, it voted for candidate bounding box by computing the similarity among edge groups. Finally, it calculated the voting scores of each candidate bounding box among multi-scale sliding windows, filterd the bounding boxes according to their scores, combined them according to their o- verlap ratio, distinguished an object according the number of combined bounding boxes, and obtained each object＇ s bounding box in the image. Experimental results show that, comparing with main methods based on bounding box, this method has the higher computation efficiency, and the recall ratio and precision are high.