Abstract: Aimed at the problems of slow extraction speed and redundancy of feature points in visual odometry algorithm, a visual odometry algorithm based on improved ORB feature is proposed. A feature region extraction strategy based on 2R criterion was designed to reduce the extraction range of feature points and speed up the algorithm. The FAST algorithm based on diagonal 8-point method was used to get more corners and improve the distribution of FAST corners, and the Quadtree algorithm based on depth was used to homogenize the feature points. Feature matching was performed to solve the camera pose. The experimental results show that the visual odometry algorithm based on improved ORB feature has better accuracy and real-time performance in camera pose estimation.