Monocular Visual Odometry Based On Inverse Perspective Mapping

The monocular vision odometry simplifies the hardware and the software as opposed to the stereo vision odometry, but it also has defect. When the vehicle is in motion, the camera's attitude changes inevitably, what lead that the method's performance degrades. To solve this problem, we proposed a monocular visual odometry based on the inverse perspective mapping (IPM). Attitude of the camera is monitored in real time by the attitude sensor when the vehicle is moving. Then the images of road surface photographed by camera became top view by using the IPM algorithm, after that, the characters of images can be calculated by the Speeded Up Robust Features (SURF) algorithm. By the random sample consensus (RANSAC) algorithm, the amounts of translation and rotation between two adjacent images can be concluded. Accordingly, the movement distance and the course of the vehicle can be worked out. In order to test the ranging accuracy of the method, both static and dynamic experiments were implemented. Static experiment showed that the average accuracy of ranging of this method achieved 1.6 parts per thousand. Dynamic experiment showed that the ranging accuracy achieved 6 parts per thousand, and the heading measurement error is less than 1.3 degrees. Therefore, the method proposed in this paper is easy to operate, time-efficient, low cost, and the accuracy of the method in ranging and heading measurement are demonstrated.