A super parallel vision system implemented with mixed analog and digital circuits

本研究的目的是开发一种新型的模拟/数字混合结构的实时图像处理视觉系统。因此，我们设计并制造了由硅视网膜、模拟超大规模集成电路（VLSI）和现场可编程门阵列组成的系统。(FPGA)。硅视网膜模拟了脊椎动物视网膜中发现的持续和短暂的反应。模拟持续响应的输出具有类似帕拉第-高斯的接受野，因此对输入图像进行平滑和对比度增强。模拟瞬态响应的输出是通过对连续图像进行减相和电阻网络平滑得到的。在室内照明下，这两个通道的输出可以在不超过几十毫秒的时间延迟内实时地从硅视网膜上交替获得。芯片的输出是偏移抑制的模拟电压，因为晶体管特性的不可控失配可以通过嵌入在每个像素电路中的采样/保持电路来补偿。来自硅视网膜的模拟输出表示具有这些通道的预处理图像，通过a /D转换器传输到FPGA。FPGA实现了边缘检测、目标跟踪和深度感知等数字图像计算。本研究开发的并行视觉系统可用于机器人视觉、多媒体等领域的各种工程应用。

The purpose of the present research is to develop a novel vision system that can carry out the real-time image processing with mixed analog/digital architecture. Accordingly, we have designed and fabricated the system consisting of a silicon retina, an analog Very Large Scale Integrated (VLSI) circuit and Field Programmable Gate Array. (FPGA). The silicon retina emulates the sustained and the transient responses found in the vertebrate retina. The output emulating sustained response possesses a Palladian-Gaussian-like receptive field and therefore carries out a smoothing and contrast-enhancement on input images. The output emulating the transient response was obtained by subtracting consecutive images that were smoothed out by the resistive network. The outputs of these two channels can be obtained alternately from the silicon retina in real time, within time delays not exceeding a few tens of ms, in indoor illumination. The outputs of the chip are offset-suppressed analog voltages since uncontrollable mismatches of transistor characteristics are compensated for with the aid of sample/hold circuits embedded in each pixel circuit. The analog outputs from the silicon retina representing the preprocessed image with these channels are transferred to the FPGA through a A/D converter. The FPGA carries out digital image computations such as edge detection, target tracking and depth perception. The parallel vision system developed in the present study is readily used in a variety of engineering applications such as the fields of robot vision and multi-media.