AirSLAM: An Efficient and Illumination-Robust Point-Line Visual SLAM System

AirSLAM introduces a hybrid visual SLAM approach that integrates deep learning for feature detection with traditional backend optimization.

Key highlights:

Unified Feature Extraction: Employs a convolutional neural network (CNN) to simultaneously extract keypoints and structural lines, enhancing feature richness.
Coupled Feature Optimization: Associates, matches, triangulates, and optimizes point and line features in a unified framework, improving pose estimation accuracy.
Lightweight Relocalization Pipeline: Introduces an efficient relocalization method that reuses the built map, utilizing keypoints, lines, and a structure graph to match query frames, ensuring robustness against long-term illumination changes.
Real-Time Performance: Achieves processing rates of 73Hz on PCs and 40Hz on embedded platforms by deploying and accelerating feature detection and matching networks using C++ and NVIDIA TensorRT.
Superior Performance: Demonstrates outperformance over state-of-the-art visual SLAM systems in illumination-challenging environments through extensive experiments.