The invention provides a visual odometer method and system based on a direct method and a storage medium. The visual odometer method comprises a brightness stabilizing step: under a high dynamic brightness changing environment, making the brightness in a scene tend to be stable with a logarithmic image enhancing method, and improving the extraction of gradient information; and an optimal convergence accelerating step: under the condition of rapid rotation and high-speed motion, generating a depth prediction surface with a plurality of remarkable gradient points and a triangular section algorithm in order to effectively give the initial depth of a point cloud, reduce the deep optimization times and accelerate the optimal convergence. The visual odometer method and system can solve the problem of incapability of normally working during great change of brightness and rapid rotation in the conventional direct method, can be applied to navigation algorithms of self-navigation robots such asunmanned aerial vehicles and unmanned vehicles, and can supply own position information in self-navigation, exploration and investigation tasks.