Chamfer-Matching (work in progress abandoned)

Warning:

If you are looking for an efficent and working C++ implementation of the Chamfer matching method, you can directly look at the Fast Directional Chamfer Matching library. This repository is only intended for versioning purpose and internal developments.

Goals:

• Detect a template image in a query image using only edge information:
  • detection at single scale.
  • detection at multiple scales.
  • detection at multiple scales and with rotation.
  • multiple detections at multiple scales and with rotation, highly cluttered background.
• Detect in a query image the most probable template image and retrieve the corresponding pose (from the same object, multiple template images at different orientations are saved with the corresponding object pose)
  • detection at single scale.
  • detection at multiple scales.
  • detection at multiple scales and with rotation.
  • multiple detections at multiple scales and with rotation, highly cluttered background.

First result:

• Template edges image:

• Query edges image:

• Detection at single scale:

References (non exhaustive):

• Chamfer Matching (contour based shape matching) - Course from University of Pennsylvania by Nicu Știurcă.
• Fast Directional Chamfer Matching (paper) - CVPR2010
• Fast Detection of Multiple Textureless 3-D Objects (paper) - ICVS2013
• Fast Detection of Multiple Textureless 3-D Objects (poster) - ICVS2013
• Contour Based Learning for Object Detection - ICCV2005
• CS664 Computer Vision - 7. Distance Transforms - Course from Cornell University
• CS664 Computer Vision - 8. Matching Binary Images - Course from Cornell University
• Comparing Images Using the Hausdorff Distance - TPAMI1993
• Hand Pose Estimation Using Hierarchical Detection - HCI2004
• Estimating 3D Hand Pose Using Hierarchical Multi-Label Classification - IVC2007
• Pedestrian Detection from a Moving Vehicle - ECCV2000
• The Chamfer System

Some available implementations in C++:

• chamfer_matching.cpp - Implementation by Marius Muja, use the old C OpenCV API.
• chamfer_matching.cpp (2) - From the Implementation by Marius Muja, some modifications added.
• chamfer_matching.cpp (3) - The original code was written by Marius Muja and later modified and prepared for integration into OpenCV by Antonella Cascitelli, Marco Di Stefano and Stefano Fabri from Univ. of Rome + fixes from Varun Agrawal.
• chamfer.cpp - From the code of Antonella Cascitelli, Marco Di Stefano and Stefano Fabri ; Rotation transform added by Aaron Damashek, John Doherty ; for the project: Detecting Guns Using Parametric Edge Matching.
• Fast Directional Chamfer Matching - Used in Edge Based Tracking library for the paper: Robust 3D visual tracking using particle filtering on the special Euclidean group: A combined approach of keypoint and edge features.

TODO:

• Allow to detect a template image with a different orientation in the query image.
• Speed-up the computation ! (Need to implement a different approach.)
• Test the robustness of the detection.
• Test the detection from multiple template images.

Ideas:

• Add a pyramidal detection (coarse-to-fine approach).
• Try to implement and use an integral distance transform image.