Vision-based odometry system and method robust to irregular illumination changes

Vision-Based Odometry Robust to Illumination Changes Using an Affine Photometric Model

This technology is a vision-based odometry system and method that minimizes photometric errors in environments with significant illumination changes by establishing an affine photometric model between keyframes and the current frame in input camera images, and performing illumination correction using planar patch-based residual vectors and Jacobian matrices.

Conventional vision-based odometry assumes constant lighting conditions, which leads to drift in estimated trajectories when sudden illumination changes occur due to camera auto-exposure, shadows, or moving clouds.

This technology proposes a method that selects planar patches using a blob detector and RANSAC, and corrects inter-frame illumination parameters by applying an affine photometric model. By using the ESM algorithm to minimize the sum of squared photometric errors, it enables robust camera pose estimation even under irregular lighting changes. It secures the reliability of position estimation in environments with severe illumination fluctuations, such as outdoor autonomous driving, drone navigation, and outdoor robotics, significantly expanding the practical application range of vision-based navigation.

‍

‍

Key Features:

• An image input unit that receives images captured by a camera, and a keyframe setting unit that handles RGB and depth images.
• A patch selection unit that selects one or more planar patches within the RGB image of a keyframe.
• An illumination model setting unit that defines the photometric model applied when transforming the current image frame.
• A configuration that obtains residual vectors using keyframes, current image frames, and model parameters, and calculates update values using a correction matrix.

‍

This invention was developed with support from the Ministry of Science, ICT and Future Planning for the "Research on Collaborative Manipulation and Network Systems for Lunar Surface Sampling and Exploration [Phase 1/Year 1]" project.