Camouflaging an object from many viewpoints

  • Published on
    16-Aug-2014

  • View
    268

  • Download
    5

DESCRIPTION

CV

Transcript

Camouflaging an Object from Many Viewpoints Andrew Owens1, Connelly Barnes2, Alex Flint3, Hanumant Singh4, and William Freeman1 1MIT CSAIL, 2University of Virginia / Adobe, 3Flyby Media, 4Woods Hole Oceanographic Inst Presenter: @miyabiarts CV CVPR 26, July 2014 Camouflage Disruptive ColorationMasquerade Optical CamouflageBackground matching Motivation Goal 3D Application: Public Art http://joshuacallaghan.com/Tianmu.htm Significance of camouflage research CV Dazzle http://cvdazzle.com/ Goal 3D Contribution 3D 37 Algorithm Background matching Two Stage Algorithm Stage1: Capture images of object Stage2: Camouflage object Stage 1: Capture Images of Object 10-25 Stage 1: Capture Images of Object SfM3D Target: Stage2: Camouflage Object Methods Nave model mean color Projection from viewpoint Random/Greedy MRF (Proposed) Interior/Boundary MRF Nave model: mean color Viewpoint 1 Viewpoint 2 Nave model: mean color Viewpoint 1 Viewpoint 2 Projection from viewpoint Viewpoint 1 Viewpoint 2 Random/Greedy Random Greedy Random Viewpoint 1 Viewpoint 2 Random Viewpoint 1 Viewpoint 2 Random/Greedy Random Greedy 70 Greedy Viewpoint 1 Viewpoint 2 Greedy Viewpoint 1 Viewpoint 2 MRF (Proposed) MRF Formulation Data Smoothing MRF (Proposed) Interior MRF Boundary MRF Data cost term Data Occlusion Viewpoint Stability Data cost term Occlusion Viewpoint Stability Interior MRF Smoothing cost term Interior MRF Viewpoint 1 Viewpoint 2 Boundary MRF Projection Boundary MRF Viewpoint 1 Viewpoint 2 Psychophysical study design Amazon Mechanical Turk Work http://camo-exp.appspot.com/game Evaluation metrics Confusion rate [s] Time to find [s] Experiment 37 Results Results MRF Greedy vs Boundary MRF Interior MRF vs Boundary MRF Results from multi viewpoints Boundary MRF https://www.youtube.com/watch?v=NNlE_hzqdss Conclusion 3D MRF

Recommended

View more >