1ms Visual Feedback System
Summary
We have developed a visual servo system with 1ms visual feedback using a massively parallel processing vision system, and realized remarkably fast and smooth target tracking.
Most conventional visual feedback systems are not fast enough to control robots by visual servoing, because their feedback rate is restricted by the frame rate of CCD cameras. In order to solve this problem we use a massively parallel processing vision system called SPE-256 in which processing elements are attached to photo-detectors. We utilize a SIMD architecture comprising of 16x16 processing elements (PEs) where each PE is 4-neighbor connected. Each PE is programmable and is capable of performing 8 bit serial arithmetic and logical functions and contains a 24 bit local memory. Therefore the system can perform a variety of image processing and visual data parallel algorithms. In the application of target tracking, our system can recognize the target and control the actuator in 1ms which is far beyond video rates. Moreover it can track the target even when it is moving quickly and unpredictably.
In following mpeg demo, we describe our visual feedback system and demonstrate high speed target tracking which can be used for various engineering applications. A one-chip fully integrated version with 64x64 PEs based on SPE architecture has been already developed.
Movies
References
- Yoshihiro Nakabo, Idaku Ishii, and Masatoshi Ishikawa: High Speed Target Tracking Using 1ms visual Feedback System, Video Proceedings of IEEE Int. Conf. Robotics and Automation (Minneapolis, 1996.4.25)/Video Proc. Abstract, p.6
- Idaku Ishii, Yoshihiro Nakabo, and Masatoshi Ishikawa: Target Tracking Algorithm for 1ms Visual Feedback System Using Massively Parallel Processing, IEEE Int. Conf. Robotics and Automation (Minneapolis, 1996.4.25)/Proc. IEEE Int. Conf. Robotics and Automation, pp.2309-2314
