Real-time motion detection is a key technology in the intelligent video surveillance and traffic video flow. Widely used methods based on monocular vision are sensitive to light and have problems of black hole and shadow. To solve these problems, we proposed a motion detection method based on binocular vision. The segmentation of move object is based on the distinctive of depth information which is obtained from two camera's parallax. Based on the results of segmentation and related depth information we proposed a novel speed detection algorithm which can adapt to a variety of space shooting angles. Experimental results show that our method obtained accurate contour and speed of the moving object which satisfactorily solved the problems brought by the monocular vision method and achieved the state-of-the-art performance.