Enhanced 3D Mapping for Mobile Robots: Post-Processing of Dense Stereo-VISLAM PointClouds

Abstract

3D mapping is crucial for mobile robotics and autonomous navigation. While 3D LiDAR systems provide high accuracy, their cost restricts deployment on budget-constrained platforms. This work proposes a low-cost VISLAM approach using stereo vision for point cloud generation and enhances map quality through post-processing. We emphasize the collection of a dedicated 3D dataset specifically designed for benchmarking and evaluating different filtering techniques. Spectral analysis and Shannon entropy are used to detect structural patterns, reduce noise, with step-by-step visualization. The method is ROS 2-compatible and suited for resource-constrained environments, enabling accurate 3D perception from visual data alone.