High-speed Vision with Massively Parallel Coprocessors
High-speed image processing at high frame rates is highly effective in real-time applications, including automation, inspection, robot control, microscope observation, and manmachine interfaces. However, the performance demonstrated thus far is not high enough to observe and control phenomena that typically require data output at high throughput rates and low latency.
Our proposed system involves three effective strategies: (A) parallelization of algorithms to minimize the amount of calculation, (B) optimal processor circuit structure for the algorithms, and (C) large scalability. We realize thesefeatures by introducing a swappable massively parallel processing module as a co-processor having specialized functionality and configuration. The developed system demonstrates a throughput of 1 kfps and a latency of several milliseconds.
- Co-processor for High-speed Moment-based Analysis of Numerous Objects [link]
- Real-time Particle Measurement/Fluid Measurement [link]
- Real-time Shape Measurement of a Moving/Deforming Object [link]
- Yoshihiro Watanabe, Takashi Komuro, and Masatoshi Ishikawa: A High-speed Vision System for Moment-based Analysis of Numerous Objects, 2007 IEEE International Conference on Image Processing(ICIP'07) (San Antonio, 2007.9.19) /Proceedings, pp. V177-V180. [PDF] *IEEE