Monocular Visual SLAM Using Edge Points for Low-Texture Environments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional visual SLAM systems face challenges in low-textured environments due to insufficient feature correspondences, leading to partial reconstruction and camera tracking failures, especially when point features are scarce or erroneous.
Innovation Solution
The proposed system employs edge points based monocular visual SLAM that utilizes edge detection and line correspondences to improve feature extraction and tracking, allowing for robust camera and structure estimation even in low-textured scenes by identifying keyframes based on rotation, feature correspondence, and spatial constraints, and performs two-view initialization and loop closure to rectify drifts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature based methods use SIFT, SURF or ORB to find point correspondences, then camera and structure estimation can be obtained by minimizing re-projection error, but the system fails when the number of points extracted is too less or erroneous especially when the amount of texture present in a scene is inadequate
Solution Approach 1:
The patent combines point features and line features into a unified SLAM framework. Line features are extracted using Hough transform and matched between frames, then integrated with point feature correspondences in the bundle adjustment optimization, allowing the system to maintain reliability in low-textured environments where point features alone are insufficient
Solution Approach 2:
The system creates a composite feature representation by merging point correspondences and line correspondences. The cost function minimizes both point re-projection errors and line re-projection errors simultaneously, creating a robust hybrid feature set that overcomes the limitations of using only point features in textureless scenes
2Adaptability or versatility
If direct SLAMs are used to be independent of feature extraction, then the system can work without texture, but it is prone to erroneous camera pose estimation due to incorrect photometric error estimation in case of change in lighting or view
Solution Approach 1:
The patent introduces line features as an intermediary geometric primitive that bridges the gap between point-based methods and direct methods. Lines provide structural constraints that are more robust to lighting changes than photometric methods, while still being extractable in low-texture environments, thus mediating between the two approaches
Solution Approach 2:
The system changes the error metric from photometric error (intensity-based) to geometric re-projection error (coordinate-based). By formulating the optimization in terms of 3D-2D line correspondences and point correspondences rather than pixel intensities, the method becomes invariant to lighting changes while maintaining accuracy in pose estimation
3Productivity
If conventional visual SLAM systems use point features for correspondence, then the system can perform camera position estimation, but it leads to partial reconstruction and stopping of camera tracking when the number of points extracted is too less
Solution Approach 1:
The patent performs preliminary line detection and line correspondence establishment before relying on point features. By pre-establishing line-based geometric constraints from the scene structure, the system creates a foundational framework that maintains tracking continuity even when point feature extraction fails in low-texture regions
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
Feature based visual simultaneous localization and mapping (SLAM) do not produce reliable camera and structure estimates due to insufficient features in a low-textured environment. Moreover, existing visual SLAMs produce partial reconstruction when the number of 3D-2D correspondences is insufficient for incremental camera estimation using bundle adjustment. Systems and methods of the present disclosure provide edge points based monocular visual SLAM that mitigates these problems. The SLAM is initialized through a validation process. A local optimization process is provided for stable pose estimation in situations where camera tracking becomes unreliable in a very low-textured challenging environment. An efficient and reliable loop closing process that uses structural properties of edges in the frames is also provided.