Adaptive Feature Regulation and Directional Consistency Optimization for Lightweight ORB-SLAM3 in Resource-Constrained Indoor Environments

Visual Simultaneous Localization and Mapping (VSLAM) is a foundational technology widely applied in robotics, autonomous driving, and augmented reality, enabling devices to navigate and construct environmental models using visual sensors. However, on resource-constrained hardware platforms, the real-time performance and robustness of VSLAM systems are severely hindered by limited computational resources, especially in indoor environments where illumination changes and dynamic scenes cause significant fluctuations in feature point quantity, undermining tracking stability and increasing backend optimization overhead. To address these challenges, we propose FCD-VSLAM, a lightweight ORB-SLAM3 algorithm integrated with feature point convergence adaptation and pose-based directional consistency checks. Specifically, we design a Feature-Point Convergence Adaptive Module that combines image enhancement, non-maximum suppression, and multi-objective optimization to dynamically determine the optimal feature point range. A Directional Consistency Check Skip Mechanism based on a motion model is introduced to reduce redundant computations during stable pose changes, and a Descriptor Optimization Method fusing directional information is proposed to improve matching accuracy. Evaluations on the EuRoC public dataset and a private dataset show that FCD-VSLAM outperforms the original ORB-SLAM3 and other lightweight SLAM methods, achieving an average 20.18% improvement in localization accuracy (ATE RMSE) on EuRoC sequences and a 12.45% reduction in per-frame processing time. This work advances the practical application of VSLAM on resource-limited devices, providing a reliable solution for real-time environmental perception in scenarios such as indoor robotics and portable augmented reality systems.The code are available at https://github.com/zcrop/FCD-SLAM.