An improved visual odometer algorithm based on ORB features is proposed for resolving the problem of large odometer errors due to the matching information redundancy of image feature points and high mismatching rate of the feature points in the Visual Simultaneous Localization and Mapping (VSLAM) visual odometer.First,this paper suggests module decomposition for the images and deletion of low-quality image modules and key points by using the gray scale change rate and the efficiency factor of subregion features,so as to resolve the problem of low extraction efficiency in the course of key point extraction and unreasonable distribution of the key points.As for the mismatching of excessive invalid feature points in the process of two-frame image matching,this paper proposes a method of deleting the feature points that are not in the image view of last frame in the image of current frame by using the camera pose of last moment.Test results show that the improved visual odometer algorithm has higher precision in camera pose estimation compared with the previous one.