Visual odometry method, system and storage medium based on direct method

Abstract

The invention provides a visual odometry method, system and medium based on the direct method, including: brightness stabilization step: in a high dynamic brightness transformation environment, using a logarithmic image enhancement method to stabilize the brightness in the scene and improve the brightness of the scene. Gradient information extraction; speed up optimization convergence steps: in the case of fast rotation and high-speed motion, use several significant gradient points and triangulation algorithm to generate depth prediction surface, effectively give the initialization depth of point cloud, reduce the number of depth optimization, and speed up Optimization converges. The invention can overcome the problem that the traditional direct method cannot work normally when the brightness changes dramatically and rotates rapidly, and can be used in the navigation algorithm of autonomous navigation robots such as unmanned aerial vehicles and unmanned vehicles, and can be used in autonomous navigation, exploration, investigation, etc. The task provides its own location information.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a robust direct method-based visual odometry method, system and storage medium. Background technique [0002] At present, it is extremely difficult for self-service robots to navigate in complex environments (such as indoors, jungles, caves, etc.), especially when the robot is moving at high speed and the surrounding environment changes (light changes, object movement, etc.). In this case, the general visual odometry is easily interrupted due to the sudden change of the environment. The visual odometry using the feature point matching method has high requirements for texture information. The visual odometer will lose its ability to work, and another direct visual odometry that directly uses pixel gray levels for matching has higher requirements for lighting, and once this type of visual odometer is interrupted once, because there is no feature matching, It is difficult t...

AI Technical Summary

Problems solved by technology

In this case, the general visual odometry is easily interrupted due to the sudden change of the environment. The visual odometry using the feature point matching method has high requirements for texture information. The visual odometer will lose its ability to work, and another direct visual odometry that directly uses pixel gray levels for matching has high requirements for lighting, and once this type of visual odometer is interrupted once, because there is no feature matching, It is difficult to restore to the state before the interruption, so the present invention improves the direct method visual odometry, and combines the relocation function of the feature point method visual odometry to propose a robust visual odometry method

Images