This technology is a structural environment feature-based visual odometry system that estimates a camera's 6-DOF trajectory by combining line information from color images with plane information from depth images. It constructs a Manhattan frame by extracting vanishing directions from line data and surface normal vectors from plane data.
Existing visual odometry techniques suffer from cumulative errors when estimating rotational movement, and methods relying solely on plane information often fail or produce inaccurate results in environments lacking sufficient planar surfaces.
By complementarily utilizing line and plane information to perceive spatial orientation, this technology first estimates rotational movement to eliminate errors, then minimizes feature point residuals based on depth availability using the Levenberg-Marquardt algorithm. This enables precise 6-DOF trajectory estimation without cumulative drift. Applicable to indoor autonomous robots, AR/VR devices, and drone navigation, it provides a cost-effective solution for high-precision localization using only a camera, without the need for expensive additional sensors.
N/A
