1000 fps Visual Servoing on the Reconfigurable Wide SIMD Processor

Visual servoing has been proven to obtain bet-ter performance than encoders at comparable cost. However, the often computationally intensive vi-sion algorithms and the ever growing demands for higher frame rate make its realization very chal-lenging. This paper demonstrated the feasibility of achieving high frame-rate visual servoing appli-cations on the wide Single-Instruction-Multiple-Data (SIMD) processors. An industrial applica-tion, Organic-Light-Emitting-Diode (OLED) sub-strate localization, is chosen as a typical example. Firstly, we proposed an improved vision pipeline for repetitive structure localization, which is more robust, and also more friendly to embedded pro-cessors and FPGA/ASIC realization. Then, a highly-efficient SIMD processor for vision servo-ing is also proposed. The number of processing el-ements (PEs) in this proposed SIMD processor is dynamically reconfigurable, which (i) enables ef-ficient processing for different vector lengths; (ii) can easily adjust to the various performance re-quirements; and (iii) saves energy consumption. For input frames of 120 × 45 resolution, the pro-posed vision pipeline can process a frame within 275 µs, sufficient to meet the throughput require-ment of 1000 fps with a latency of less than 1 ms for the whole vision servoing system. Compared to the reference realization on MicroBlaze, the proposed SIMD processor achieves a 18 × perfor-mance improvement.