Transcript Week 10 7/16/14 - UCF Center for Research in Computer Vision

Week 9
7/16/14
Amari Lewis
Aidean Sharghi
Light field dataset
• Using the depth estimate provided by the Lytro compatible viewer software.
• See if we can use this information to increase object recognition.
• The white pixels do not
have depth value- due
to occlusion, distance,
surface angle or
material.
• The darker the color
(black or grey), the
more accurate depth
perception)
bike
vehicle
building
RGB-D cameras
• Sensing systems that capture RGB images along with per-pixel depth
information.
• The white pixels do not have depth value- due to occlusion, distance, surface
angle or material.
• The darker the color (black), the more accurate depth perception)
Studying the depth information…
• RGBD – OBJ CALSSIFIATION
• The process(RGB-D object recognition and detection):
utilize sliding window detectors trained from object views to assign class probabilities to pixels
in every RGB-D frame.
• Our ultimate goal is to find out a way that we can incorporate the use of the
depth estimation from light field images for object recognition.
References
• Holistic Scene Understanding for 3D Object Detection with RGBD camera
authors: Dahua Lin Sanja Fidler Raquel Urtasun
• RGB-D Object Recognition and Detection – Artificial intelligence University of
Washington
• Depth from Combining Defocus and Correspondence Using Light-Field
Cameras-University of California, Berkeley
Authors: Michael Tao1, Sunil Hadap2, Jitendra Malik1, and Ravi Ramamoorthi
• RGB-D Mapping: Using Depth Cameras for Dense 3D Modeling of Indoor
Environments authors: Peter Henry, Michael Krainin, Evan Herbst, Xiaofeng Ren,
Dieter Fox