Visual servoing applies image sensors instead of mechanical encoders for position acquisition. It can achieve higher resolutions than mechanical encoders at comparable cost. However, visual servoing is computation intensive. Special purpose hardware is often required. This work performs a case study on organic light emitting diode (OLED) manufacturing, a typical industrial application of machine vision. We optimize the vision processing algorithm and implement it on a field-programmable gate array (FPGA). Timing analysis is performed to identify the bottlenecks of the implementation. This work identifies the exposure time and the camera interface as the bottlenecks of the high frame rate visual servoing (above 1000 frames per sec- ond). A tracking task is simulated to evaluate the performance of the visual servoing system. We explore the tradeoffs between frame rate, delay, and performance. This work has special emphasis on the time predictability of the visual servoing system, which is a critical requirement for industrial applications. Our implementation exploits the synergy of algorithm, architecture, and interface to guarantee the predictability of the whole system.
|Title of host publication||Proceedings of the 12th IAPR Conference on Machine Vision Applications [MVA'11], June 13-15, 2011, Nara, Japan|
|Publication status||Published - 2011|
Ye, Z., He, Y., Pieters, R. S., Mesman, B., Corporaal, H., & Jonker, P. P. (2011). Bottlenecks and tradeoffs in high frame rate visual servoing : a case study. In Proceedings of the 12th IAPR Conference on Machine Vision Applications [MVA'11], June 13-15, 2011, Nara, Japan (pp. 55-58)