june 3, 2013 apaar sadhwani and jason su
DESCRIPTION
June 3, 2013 Apaar Sadhwani and Jason Su. Motivation. Goals. Methods & Approach. Challenges. Pipeline. Table Detection. Pipeline. Table Detection. Pipeline. Table Detection. Pipeline. Table Detection. Pipeline. Table Detection. Pipeline. Ball Segmentation. Pipeline. - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/1.jpg)
Mobile Pool Table Analysis
June 3, 2013Apaar Sadhwani and Jason Su
![Page 2: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/2.jpg)
Motivation
Novices have difficulty with shot selection in billiards
Players have no practical way of analyzing their performance
Mobile devices are ubiquitous and turn-based play makes recording possible
![Page 3: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/3.jpg)
Goals
Capture table state from a single photo
Populate a virtual game table with the state
Shot selection and guidance with rendered graphics
![Page 4: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/4.jpg)
Methods & Approach
Apply strong constraints on the problem• Table and ball size is standardized• Planar scene• A single dominant hue near the center of the
photo
OpenCV and OpenGL ES 2.0
![Page 5: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/5.jpg)
Challenges
View from any angle
Ball occlusion
Table occlusion due to FOV or objects
![Page 6: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/6.jpg)
Pipeline
Table Detection
![Page 7: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/7.jpg)
Pipeline
Table Detection
Surface
• Hue based seg./cropping• Phase-unwrapping and
equalization
![Page 8: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/8.jpg)
Pipeline
Table Detection
Surface
• Hue based seg./cropping• Phase-unwrapping and
equalization
![Page 9: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/9.jpg)
Pipeline
Table Detection
Surface
• Hue based seg./cropping• Phase-unwrapping and
equalization
Edges
• Hough transform• Rejection using surface-
based metric
![Page 10: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/10.jpg)
Pipeline
Table Detection
Surface
• Hue based seg./cropping• Phase-unwrapping and
equalization
Edges
• Hough transform• Rejection using surface-
based metric
Orientation
• Homography• Hypothesis testing
![Page 11: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/11.jpg)
Pipeline
Ball Segmentation
Blobs
• High-resolution processing• Connected component
labeling
![Page 12: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/12.jpg)
Pipeline
Ball Segmentation
Blobs
• High-resolution processing• Connected component
labeling
Positioning
• Circular Hough transform• Radius range from
homography
![Page 13: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/13.jpg)
Pipeline
Ball Segmentation
Blobs
• High-resolution processing• Connected component
labeling
Positioning
• Circular Hough transform• Radius range from
homography
Rejection
• Collision physics• Pixel radius constraints
![Page 14: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/14.jpg)
![Page 15: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/15.jpg)
Pipeline
Table Detection
Surface
• Hue based seg./cropping• Phase-unwrapping and
equalization
Edges
• Hough transform• Rejection using surface-
based metric
Orientation
• Homography• Hypothesis testing
Ball Segmentation
Blobs
• High-resolution processing• Connected component
labeling
Positioning
• Circular Hough transform• Radius range from
homography
Rejection
• Collision physics• Pixel radius constraints
![Page 16: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/16.jpg)
Display
![Page 17: June 3, 2013 Apaar Sadhwani and Jason Su](https://reader030.vdocuments.mx/reader030/viewer/2022033105/568164ba550346895dd6c5b3/html5/thumbnails/17.jpg)
Discussion
Pipeline is heavily dependent on surface segmentation• Plays a critical role in the edge rejection metric• Also it is itself a major edge for finding lines
Balls localization• Rejection needs work• Occlusion is difficult
Video tracking