Visual odometer implementation method based on semi-direct method

The invention relates to a visual odometer implementation method based on a semi-direct method, and belongs to the technical field of visual space positioning. For a dynamic scene, in order to utilize most useful image information and reduce a non-convexity effect of minimizing a luminosity error, a mixed point extraction method combining a high gradient point and an ORB angular point is adopted, and a pose obtained by minimizing a re-projection error is used as an initial value to minimize the luminosity error, so that a final camera pose is obtained. In addition, dynamic features are effectively eliminated in the processes of point selection and luminosity error generation by using semantic information, so that robust and accurate pose estimation is realized in a dynamic scene. The method can effectively reduce the adverse effect of a dynamic object on the positioning precision and robustness of a visual odometer system, and improves the accuracy of camera attitude estimation while retaining the robustness