Object Recognition for Amazon Picking Challenge Patrick...
Transcript of Object Recognition for Amazon Picking Challenge Patrick...
Object Recognition for Amazon Picking Challenge Patrick Greene
Introduction (Amazon Picking Challenge)
Introduction (Problem Setting)
Disscussion of Previous Works (3D)
example point cloud taken by a
kniect
Lai, Kevin, et al. "Sparse distance learning for object recognition combining rgb and depth information." Robotics and Automation (ICRA), 2011 IEEE International Conference on. IEEE, 2011.
example point cloud taken by a
kniect
Disscussion of Previous Work (Classifiers)
Savarese, Silvio, and Li Fei-Fei. "3D generic object categorization, localization and pose estimation." Computer Vision, 2007. ICCV 2007. IEEE 11th International Conference on. IEEE, 2007.
Discussion of Previous Work (Good old SIFT)
Lowe, David G. "Object recognition from local scale-invariant features."Computer vision, 1999. The proceedings of the seventh IEEE international conference on. Vol. 2. Ieee, 1999.
Detailed Description (My Problem/Assumptions)
Detailed Description (Assumptions)
Detailed Description (Mechanical Assumptions)
Detailed Description (Plane to Plane Recognition)
Under the assumption that every object is componed entirely of planes, all that must be done is search for all the planes that compose an object.
Which Descriptor/Extractor
Detailed Decription (Methodology)
Fund. Matrix RANSAC FLANN Matcher
Detailed Description (Redaction)
Detailed Description (Dove Phenomenon)
Detailed Description (Unlikely Orientations aka. Probability)
Experiments and Results (Descriptor/Extractor)
Experiments and Results (Full Test Confusion Matrix)
Objects True Positive
True Negative
Bad Positive
False Negative
Tom Sawyer
X
Glue X X
Joke Book X
Crayola X
Outet Plugs X
Spark Plugs X
Color Cups X
Experiments and Results (Repeated Object Test)
Experiments and Results (Classifier)
Discussion (Failures Study)
Discussion (Failures Study)
Disscusion (Failures Study)
Discussion (was never gonna work)
This slide is not blank (Training Image Back of Notecards)
Disscussion (Gotta Go Fast)
Full processing aka. failure time <= 1minute
That means that 1 image at a time we could fail 20times. With parralel processing of images this should be fine.
Discussion (Potential Changes)
Addition of simple blob detector
Addition of color based detector
Changes to lighting, camera tpye/orientation
Questions/Suggestions