A Visual-Inertial SLAM Method for Robots on Resource-Constrained Platforms Using a Low-Cost Binocular Camera

Visual Simultaneous Localization and Mapping (SLAM) algorithms relying on cameras have gained significant attention with advancements in computer vision. However, existing approaches often depend on high-performance hardware, limiting their applicability in resource-constrained scenarios. This study addresses indoor planar motion applications, such as service robots and automated guided vehicles (AGV), by proposing a cost-effective Visual-Inertial SLAM(VI-SLAM) framework tailored for low-performance platforms using a low-cost binocular camera and an inertial measurement unit (IMU). A localization module is developed to enhance feature stability through inertial prediction and reprojection cross validation, using grayscale images, depth images, and IMU data. A map correction module is introduced to integrate backend optimized poses, loop closure detection, and depth information, enabling the generation of globally consistent probability grid maps via submap stitching and multi-state depth image conversion. Comprehensive evaluations are conducted on public datasets and real-world environments. The results demonstrate the accuracy, robustness, and feasibility of our approach for low-performance and low-cost devices, highlighting its potential for commercialization and practical applications.