Robust 3D Surveillance
Abstract
The main objective of this proposal is to design a real-time computational framework for distinguishing between real 3D video streams captured by RGB-D cameras and the reconstructed rendering of such 3D video sequences designed to circumvent 3D surveillance systems. Previously, a basic framework that uses a skeleton-based prediction mechanism for manipulating the 3D scenes from a single RGB-D camera was developed. A user study was conducted to visually inspect the manipulated depth, color and 3D video streams and check whether humans are able to identify/recognize the manipulations. This study showed that it is extremely difficult for humans to detect the manipulations. To overcome this difficulty the proposed approach is to first, design a method that can generate manipulated 3D scenes with minimum distortion by minimizing the artifacts that arise due to the manipulations in the 3D scenes. This method, to be called DISPOSE (DIstortion Score based POse mesh Selection), will use a database of 3D meshes to select the mesh that would be appropriate for the desired output animation with highest possible quality. Then, algorithms will be devised to detect differences in depth, color, and segmentation between the DISPOSE manipulated streams and the real video streams acquired from multiple RGB-D cameras.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 12, 2017
- Source ID
- W911NF1610163
Entities
People
- Balakrishnan Prabhakaran
Organizations
- Army Contracting Command
- United States Army
- University of Texas at Dallas