How taeyoung96/visual-odometry-tutorial works
Project type and purpose
• This is a Python-based computer-vision tutorial repository (“Visual-odometry-tutorial”) that demonstrates basic stereo/monocular vision techniques (feature matching, epipolar geometry) and a lightweight visual-odometry (VO) pipeline using two popular datasets (KITTI and TUM). It is neither a web app nor a microservice: it’s essentially a CLI-driven pipeline with modular scripts.
File structure
• Top-level directories and their roles:
• cv_basics/ – contains two demo scripts (feature_matching.py, epipolar.py) for low-level CV visualization.
• visual-odometry/ – contains vo.py (main VO pipeline) plus pose_evaluation_utils.py for trajectory evaluation and plotting.
• data/ – holds minimal KITTI and TUM dataset subsets.
• result/ – stores output images (feature matching, epipolar lines, VO trajectories).
• slides/ – tutorial slides (VO Tutorial.pdf).
• requirements.txt – Python dependencies.
• Patterns & config:
• Each module is a stage in a linear data-processing pipeline.
• vo.py is parameterized via CLI arguments (--data_dir_root, --dataset_type, --len_trajMap).
README for additional insights
• Confirms the pipeline stages, example commands, and that results are purely estimated (no optimization).
• Specifies environment (Anaconda, Python 3.8), dependencies (OpenCV, NumPy, Matplotlib, etc.), and dataset layout expectations.
a. Main components to represent:
1. CLI Interface / Runner – parses arguments, dispatches to selected script.
2. Data Loader – reads images and ground-truth files from data/kitti-odom or data/tum.
3. CV Baseline Module – two sub-components:
• Feature Matching (cv_basics/feature_matching.py)
• Epipolar Geometry (cv_basics/epipolar.py)
4. VO Pipeline Module (visual-odometry/vo.py) – comprising:
• Feature Extraction & Matching
• Pose Estimation (PnP or Essential‐matrix decomposition)
• Trajectory Accumulation
5. Evaluation & Visualization (pose_evaluation_utils.py + Matplotlib) – generates Trajectory.png, feature_matching.png, epipolar.png.
6. Result Store – writes image files to result/ and .txt outputs.
b. Relationships & data flow
• From Data Loader to CV Baseline Module (feature & epipolar) and/or VO Pipeline.
c. Architectural patterns & design principles
• Linear processing pipeline (sequential stages).
• Modular separation (cv_basics vs. visual-odometry).
• CLI-driven configuration for reproducibility.
d. Technologies & frameworks
• Python 3.8, OpenCV (cv2), NumPy, Matplotlib.
• Anaconda environment.
• Datasets hosted externally (KITTI, TUM).