In order to improve the real-time performance of the traditional stereo matching algorithm based on census transform, which has low matching accuracy in weak texture regions and is easy to be affected by noise points, a novel stereo matching algorithm based on Census transform and adaptive window is proposed. In the cost calculation stage, the window size is adaptively matched according to the strength of the regional texture, three kinds of state information are used for Census transformation to calculate the initial cost, which improves the matching accuracy of single pixel and reduces the consumption time. Then, in the cost aggregation stage, a low time complexity guided filter is used to solve the problem of low accuracy caused by low cost discrimination of single pixel matching. Finally, the left-right consistency detection principle is used to reduce the outliers, and the final disparity map is obtained. The proposed algorithm is tested using Middlebury platform standard images. The experimental results show that the average error matching rate of the proposed algorithm is 5.51%, and the matching accuracy is improved to some extent. The average time-consuming is shortened by 36.60% compared with the traditional Census algorithm, and the real-time performance of the algorithm is improved.