VSLAM: Stereo Visual SLAM Pipeline
Visual Simultaneous Localization and Mapping (VSLAM) is a fundamental capability for autonomous robots and vehicles. This project implements a complete stereo visual SLAM pipeline in pure Python, designed for clarity and extensibility while remaining compatible with standard benchmarks.
View the VSLAM GitHub Repository
Overview
The pipeline takes stereo image pairs and produces a 3D trajectory estimate of the camera’s path through the environment. It combines classical computer vision techniques into a cohesive system that can be used for research, education, and prototyping.
Pipeline Components
Feature Detection & Tracking – Detects keypoints across stereo frames and tracks them temporally using optical flow, providing robust correspondences for motion estimation.
Stereo Matching – Matches features between left and right camera images to triangulate 3D landmark positions using known camera geometry.
Motion Estimation – Estimates camera pose changes between frames using PnP (Perspective-n-Point) for 2D-3D correspondences and ICP (Iterative Closest Point) for 3D-3D alignment.
Bundle Adjustment – Jointly optimizes camera poses and 3D landmark positions over a sliding window to minimize reprojection error and improve trajectory consistency.
Technical Stack
- Python with NumPy for core computation
- OpenCV for feature detection, matching, and image processing
- KITTI benchmark compatible input/output formats
- pytest test suite for validation of individual components
Key Design Decisions
The pipeline is implemented entirely in Python to maximize readability and ease of modification. While this trades off raw performance compared to C++ implementations, it makes the codebase accessible for researchers wanting to experiment with individual components or students learning about SLAM systems.
Requirements
- Python 3.8+
- OpenCV
- NumPy
- SciPy (for optimization)
License
MIT License