A2PM-VINS: A Visual-Inertial SLAM Method Based on Area-to-Point Matching.

Area of Science:

• Robotics
• Computer Vision
• Simultaneous Localization and Mapping (SLAM)

Background:

• Visual-inertial SLAM (VI-SLAM) localization accuracy is vital for autonomous systems.
• Degraded environments (low illumination, repetitive/weak textures) challenge traditional VI-SLAM front-end feature matching, causing sparse features, mismatches, and unstable state estimation.

Purpose of the Study:

• To propose a novel Area-to-Point Matching Visual-Inertial SLAM (A2PM-VINS) method.
• To enhance the reliability and robustness of VI-SLAM in challenging, degraded environments.

Main Methods:

• Introduced Area-to-Point hierarchical matching and a kinematic temporal inheritance mechanism for improved matching reliability and track continuity.
• Developed an Anchor-Explorer feature selection strategy to prioritize geometrically valuable features for back-end optimization.
• Incorporated a Sub-Window Consistency (SWC) weighting strategy in the back end to mitigate geometrically deceptive observations.

Main Results:

• A2PM-VINS demonstrated superior or competitive localization accuracy on challenging sequences within the EuRoC MAV dataset.
• Achieved low absolute trajectory errors (0.0983 m on MH_04, 0.1191 m on MH_05).
• Maintained stable tracking on V2_02, outperforming VINS-Fusion in a failure case.

Conclusions:

• The proposed A2PM-VINS method significantly improves the robustness of visual-inertial state estimation in complex, degraded environments.
• A2PM-VINS offers a more reliable solution for VI-SLAM applications operating under adverse conditions.