Method and Apparatus for Estimating Self-Position of Autonomous Driving Device Using Distorted Images

Method for Estimating Self-Position by Projecting Distorted Images onto a Cubemap

This technology is an image processing algorithm that enhances the accuracy of feature-based Simultaneous Localization and Mapping (SLAM) by projecting wide-angle camera distorted images onto a cubemap, then cropping and performing perspective transformation on specific viewpoints (front and floor).

When using camera-based SLAM, noise or distortion caused by changes in lighting leads to cumulative positioning errors over time, which reduces map accuracy.

This technology corrects distortion and ensures feature consistency by converting distorted images into a cubemap and performing perspective transformation on cropped areas, such as the floor, to improve positioning accuracy.

‍

‍

Key Features:

• Receiving a distorted image captured by a wide-angle image sensor of an autonomous driving device
• Mapping the distorted image onto a cubemap to generate a cubemap image associated with multiple viewpoints, including a first viewpoint and a second viewpoint
• Performing a perspective transformation on the first mapped image from the first viewpoint relative to the second mapped image from the second viewpoint
• Estimating the self-position of the autonomous driving device based on the perspective-transformed first mapped image and the second mapped image

‍