Algorithms for Enabling Point-and-Shoot 3D Camera
Computing & Wireless : Computing Methods
Available for licensing
- Sriram Vishwanath, Ph.D. , Electrical and Computer Engineering
- Chris Slaughter , University of Texas at Austin
3D cameras are the next major step in the development of visual media. However, enabling 3D content requires considerable algorithmic sophistication and mathematical insights. For these reasons, 3D cameras do not exist presently. Current technologies attempt to perform 3D capture using expensive hardware (such as LIDARs) and/or labor-intensive stitching of measurements, thus are not real-time and, in most cases, not true 3D in their output.
This invention is a design of algorithms for enabling point-and-shoot 3D cameras. It creates a highly accurate 3D model of an observed scene by combining multiple views from a moving camera. The output of the system is a point cloud or mesh model. Models can be captured at arbitrary scales varying from small objects to entire buildings. The visual fidelity of produced models is comparable to that of a photograph when rendered using conventional graphics rendering. Despite offering fine-scale accuracies, the mapping results are globally consistent even at large scales.
- Computational efficiency: software execution is several orders of magnitude faster than current state-of-art
- Accuracy: method obtains super resolution 3D models
- Flexibility: method works with a variety of sensor modalities
- Scaling: captured models can scale to arbitrary sizes without introducing artifacts
- High-fidelity 3D images using a moving camera
- Sophisticated algorithms render images that are high-resolution and highly scalable.
- No expensive hardware is necessary.
3D imaging, visual effects, spatial acquisition, real estate, architecture, augmented reality