Utilizing off-the-shelf hardware for advanced 3D vision
Robots and autonomous vehicles need to identify objects to interact with their environment and avoid obstacles. Distinguishing a wall from a door or seeing an object placed against a similarly-colored background is simple for a human, but can be difficult for a machine. Various techniques have been developed to solve this problem, but they often involve expensive sensor hardware and require a large amount of computational power.
A computational technique developed by researchers at the University of Pennsylvania can segment images to easily identify objects using simple RGB-D cameras. A color image and depth data are gathered using the camera, and that data is then processed to find objects. From the color data, edges and discontinuities can be identified to locate distinct items. The depth data is used to find which objects stand out in 3D space. The segmentation from the color and depth data is matched together to identify discrete objects in a fast and computationally efficient manner. This method provides simple and cost-effective 3D computer vision for robotics using sensors such as the Microsoft Kinect.
- Uses low-cost RGB-D sensors
- Fast and efficient segmentation based on color and depth data
- Ideal for mobile autonomous applications
Stage of Development:
Proof of concept prototype
United States Patent 8,867,793
Taylor, Camillo J., and Anthony Cowley. Segmentation and analysis of RBG-D data RSS 2011 Workshop on RGB-D Cameras. Vol. 90. 2011. (Graphic featured referenced from this paper)
Docket # W5413