Method and Apparatus for SLAM Optimization Using Point-Line Features

SLAM Optimization Method Enhancing Precision via Point-Line Feature Fusion and Vanishing Points

This technology defines a 'point-line feature' by combining point and line features extracted from images. It secures high-precision localization and robust map generation by integrating vanishing point-based line grouping and reprojection error minimization into a sliding window-based SLAM optimization process.

Existing SLAM technologies based solely on point or line features often suffer from localization and mapping errors in changing environments or specific geometric structures (such as pure rotation). Furthermore, they face technical limitations where the optimization process can get trapped in local minima, leading to reduced convergence speed and accuracy.

This technology extracts point features from edge change points in an image and line features—including start and end points—from multiple edges. It then generates point-line features by pairing each point feature with its nearest line feature. By minimizing the distance between reprojection points and lines, and grouping lines that share the same vanishing point for use in optimization calculations, the system significantly improves overall accuracy.

‍

‍

Key Features:

• Receiving images containing multiple frames and extracting multiple point features from locations where edge directions change.
• Extracting multiple line features, including start and end points, from multiple edges within the image.
• Determining the line feature with the shortest perpendicular distance from a target point feature as the target line feature, and performing SLAM optimization using the combined point-line feature.
• Grouping first and second line features that share the same vanishing point beyond a threshold number of times within a sliding window.

‍