FPGA-based embedded visual servoing platform for quick response visual servoing

Abstract

The response speed of visual servoing depends on the computation platform's data transfer and computation capability. In this work, we proposed an embedded FPGA-based visual servoing platform (VSP) to meet the requirements on data transfer and computation capability for quick response visual servoing tasks. To address the data transfer capability issue, we adopted coprocessor-like system architecture, which provided direct connections between vision processing IPs and peripheral modules. To address the computation capability issue, we exploit as much parallelism within vision processing algorithms as possible to ensure high visual information update frequency. Two visual servoing tasks that required quick response, a ball catching and a high speed target shooting tasks, were chosen to verify the applicability and effectiveness of the VSP. In the ball catching case, we achieved 85% catch success rate and 58Hz visual update frequency. In the high speed target shooting case, we achieved 91% average hit rate and 402 Hz visual update frequency.