Robot localization based on visual odometry

With the development of computer vision technology, the positioning technology based on vision sensor had drawn more and more attentions. There were two kinds of visual odometry technologies through the image information for motion estimation to obtain the attitude information, one of them was based on the feature points and the other one was the direct method without the feature points. In recent years, many scholars approached various of methods and visual odometry algorithm based on the image data but no one is perfect. KinectV1, as a high-performance RGB-D sensor, could capture both color and depth images. The evaluation about two kinds of visual odometry technologies based on KinectV1 sensor was carried out. A summary and analysis for the robustness and accuracy problem was studied and researched. The results of evaluation showed that the method based on feature points could be applied to the environment riched in features, and the direct method is more robust in the environment of visual feature degradation.