To compute the depth information using image processing is one of key techniques in the binocular stereo camera.Stereo vision image matching principle is analyzed and a new method of realizing real-time stereo image matching algorithm based on FPGA is proposed to decrease the matching time.In order to achieve regional stereo matching，based on the Census transform，the similarity are compared through the gray level of the pixel in the neighborhood；Multiple window correlation matching method are used to improve the quality of image when depth does not match on the left-right consistency constraints，thus，the matching accuracy is improved.Real-time image matching is implemented by using FPGA pipelining and parallel processing technology.The experimental results demonstrate that the method has higher throughput rate and processing speed，the real-time processing of 1024×1024 gray image with 30fps are realized at the frequency of 97.3MHz.