TRACLabs Inc. engineers have decades of experience in developing and fielding computer vision systems for robot applications. These vision systems have been used to track people, recognize their gestures, identify objects, return the six degree-of-freedom pose of objects and perform terrain analysis for obstacle avoidance and path planning. Our computer vision software runs in real-time using range images from LIDAR or stereo.
Object recognition is one of the most difficult computational problems for robots. The nearly infinite variety of objects in all kinds of poses makes recognizing objects from camera images difficult. TRACLabs Inc. researchers are tackling this problem by applying a sequence of specialized processing filters to this 3D data to extract object type, location and orientation.
- Eric Huber and Ferid Gharsalli, “Implementation of 3D, LIDAR-Based, Vision to Improve the Safety of Grading and Excavation Machines,” in Proceedings of the 11th International Conference on Engineering, Science, Construction, and Operations in Challenging Environments, 2008.
- Eric Huber and Kenneth Baker, “Using a Hybrid of Silhouette and Range Templates for Real-time Pose Estimation“, in Proceedings of the 2004 IEEE Intnl. Conf on Robotics and Automation, New Orleans, LA, 1652- 1657 Vol.2, April 2004
- Eric Huber and David Kortenkamp A Behavior-based Approach to Active Stereo Vision for Mobile Robots, in Engineering Applications of Artificial Intelligence, Vol. 11, pp. 229-243, 1998.