In traditional monocular visual odometries, there are many false match points, low match accuracy, large amount of computation during the feature extraction process of the traditional monocular visual odometry. This paper presents an improved monocular visual odometry model based on the SURF algorithm. The feature points between two adjacent frames are detected and matched with the SURF algorithm. The RANSAC algorithm is applied to remove the error feature points. Then, the rotation matrix R and shift vector T between two adjacent frames are calculated to accomplish the motion estimation. The experiment results show that the computing speed is accelerated by 11.2% and 10.38% in the curve motion and the straight motion, respectively, with the proposed visual odometry estimating model.