Occlusion Reasoning for Object Detection under Arbitrary Viewpoint CVPR 2012 Edward Hsiao and Martial Herbert Presenter : Byungju Kim.

Occlusion Reasoning for Object De-tection under Arbitrary Viewpoint

CVPR 2012Edward Hsiao and Martial Herbert

Presenter : Byungju Kim

Overview

• Problem setting• Texture-less Object Detector• Main Idea• Experiment result• Conclusion

Assumption

• Arbitrary viewpoint• Household environment– Kitchen

• Occluder is on same support plane

Why occlusion problem?

Let’s find scissors in the pic-tures

Texture-less Object

Texture-less Object Detector

• LINE2D

Texture-less Object Detector

• LINE2D

Scoring

Z : detection windowN : number of points = 0 if the point i is occluded 1 if the point i is visible

Penalty term

Occlusion prior

𝑉 𝑖

Occlusion Conditional Likelihood

Scoring

Experiment Result

Experiment Result

Conclusion

• Calibration for multi-view is necessary• Not scale-invariant

• Idea is adaptable for different environ-ment

• Textured object could perform better

Thank you