一种基于改进块匹配算法的运动车辆检测

Moving vehicle object detection based on improved blocking matching algorithm

针对传统块匹配算法计算量大、光照变化较敏感和实时性差的缺陷,本文对块匹配算法进行了相关研究,提出了改进算法。首先把两幅彩色图像转化为灰度图像,并通过概率松弛标记算法获得边缘图像,计算相邻两幅图像的差分图像得到运动区域,然后将差分图像与边缘检测图像相与得到运动区域的边缘信息,再进行匹配得到位移矢量场并对其进行矢量中值滤波,最后通过顺序区域增长将运动车辆分割出来。实验结果表明:本文方法相对于传统方法平均检测时间降低约58ms,而平均检测率提高了约6.7%。这种算法鲁棒性强、实时性好。

For the imperfection of traditional block matching algorithm requiring large amount of computation,being more sensitive to light changes and having poor real-time,this paper conducts some related study about block-matching algorithm and proposes an improved algorithm.In this paper,block-matching algorithm is studied and an improved algorithm is proposed.First,two color images are converted to gray images,and the edge image is obtained by probabilistic relaxation labeling algorithm,to get the motion region by calculation difference image of two adjacent images.Then the edge information of moving region can be obtained by AND operation using the difference image with the edge detection image,and get the displacement vector by match operation,the vector median filtering is carried out.Finally,the moving vehicles can be obtained by region growing.Experimental results indicate that the average detection time of this method is reduced by about 58 ms compared with the traditional method,while the average detection rate is increased by about 6.7%.This algorithm has strong robustness,good real-time performance.