1000 fps visual servoing on the reconfigurable wide SIMD processor

Visual servoing has been proven to obtain better performance than encoders at comparable cost. However, the often computationally intensive vision algorithms and the ever growing demands for higher frame rate make its realization very challenging. This paper demonstrated the feasibility of achieving high frame-rate visual servoing applications on the wide Single-Instruction-Multiple-Data (SIMD) processors. An industrial application, Organic-Light-Emitting-Diode (OLED) substrate 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 processors and FPGA/ASIC realization. Then, a highly-efficient SIMD processor for vision servoing is also proposed. The number of processing elements (PEs) in this proposed SIMD processor is dynamically reconfigurable, which (i) enables efficient processing for different vector lengths; (ii) can easily adjust to the various performance requirements; and (iii) saves energy consumption. For input frames of 120 x 45 resolution, the proposed vision pipeline can process a frame within 275 mus, sufficient to meet the throughput requirement 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 18x performance improvement.