How mayankm96/stereo-odometry-soft works
Project type and purpose
• This is an academic/open‐source MATLAB implementation of a stereo‐odometry algorithm (“SOFT”) that processes synchronized left/right image pairs over time to estimate camera motion.
• It is not a client-server or full-stack app but rather a data‐processing pipeline implemented as MATLAB scripts and functions, orchestrated by main.m and visualSOFT.m.
File structure
• Top-level:
– README.md, LICENSE.md, .gitignore
– code/ directory containing config, functions, main scripts
– data/ directory holding input images and calibration files
– images/ directory holding documentation figures
• code/config:
– configFile.m defines camera parameters, bucket sizes, thresholds, etc.
• code/functions:
– featureProcessing: filters (blob, checkerboard, Sobel), descriptor computation, non-max suppression
– featureMatching: match SAD, circular matching, NCC verification, outlier removal
– featureSelection: bucketFeatures (uniform distribution)
– triangulation: linearTriangulation, cross2Matrix
– updateMotion: updateMotionP3P (P3P + RANSAC)
– utils: createCamProjectionMatrices, plotting/visualization helpers
• code/main.m and code/visualSOFT.m orchestrate the end-to-end pipeline and plotting.
README for additional insights
• The README outlines a 5‐stage pipeline: feature processing → feature matching → feature selection → egomotion estimation (triangulation + P3P) → visualization.
• It calls out MATLAB toolboxes (Parallel Processing, Computer Vision) as external dependencies.
• Input Data Loader
• Configuration Module
• Feature Processing
• Feature Matching
• Feature Selection
• Triangulation & Motion Estimation
• Visualization / Output
• External MATLAB Toolboxes (annotated as “Parallel Processing Toolbox” and “Computer Vision Toolbox”)
b. Relationships and data flow:
c. Architectural patterns/design principles:
• Pipeline pattern: sequential processing of image data.
• Modular design: each stage is a separate folder of functions.
• Reuse of utils (projection matrix creation, plotting) across modules.
• Parameter‐driven behavior (configFile.m).
d. Technologies & frameworks:
• MATLAB R2018a
• MATLAB toolboxes as external services
• Algorithmic modules implemented as .m functions grouped by responsibility
3. Next, a rectangle “Feature Matching” with sublabels: SAD matching, circular strategy, NCC outlier removal.
4. Then “Feature Selection” (bucketFeatures).
5. Then “Triangulation & P3P ‑ Motion Estimation” grouping code/functions/triangulation and updateMotion.
6. Finally, a “Visualization” block calling visualSOFT and utils plotting functions.
Emphasizing essential architectural elements
• Parameter flow from configFile into each stage.
• Pipeline ordering and key data transformations (raw images → keypoints → matches → 3D points → pose).