This paper is dedicated to the research of head detection method based on binocular stereo vision. In the first stage, we correct the acquired images, and in the second stage calculate the target parallax in binocular vision. According to the feature of the head contour, different depth distribution of person and the scene in depth map by binocular stereo image, the relationship between head's geometric features and depth information can be obtained. And then, the method of depth stratification to obtain the depth layers of head information can be used. The second part is to regard the geometric features of contour as the decision condition of round head, and then combined with the stereo matching algorithms of binocular vision to recognize the target under the binocular vision. Finally, whether the contour is the head or not can be distinguished. Experimental results show that the algorithm can be adapted to detect the head effectively in complex situation and has high accuracy.