ResistiveNetwork Introduction

Background

Visual Feedback System Using Parallel Vision.

For realizing visual feedback control, development of a high-speed vision system with low cost is essential.

The projection of the working coordinate system on the imager is not Cartesian but nonlinear (in other words, nonuniform).

For feedback control using visual feedback, direct detection of feature values in nonlinear coordinate systems is useful.

Parallel Vision System

Paralell input, parallel calculation

The formar half of this figure is realized as vision chips. They can compute early-vision-like processing in parallel. However, to close the feedback loop, feature values such as centroids must be extracted. A circuit for this purpose is proposed here.

About Vision Chips

Detecotors and processing elements for parallel processing are fabricated on the same chip and processing elements are connected to the circuit in parallel. Many types of vision chips are proposed in such uses as edge detection, calculation of optical flow, reconstruction of surfaces, and binocular vision. ( A survey of vision chips)

Tey are very high speed。For example

Edge detection 1ms (MIT Vision Chip, 1991)

In addition, Ishikawa et al. developed an array of programmable processing elements, which, although a macromodel of a vision chip, can perform such functions as edge detection, thinning, and detection of moving objects. Most processes are completed within one millisecond.

A Circuit for Detecting the Centroid of Two-dimensional patern on Nonlinear Coordinates

Request to the circuit:

To take advantage of high-speed early-vision-like processing, the centroid-extraction circuit must receive outputs of the former circuit in parallel and calculates feature values in parallel.

In visual feedback control, it is useful to obtain feature values in relation to working coordinates instead of image coordinates.

A circuit which extract the centroid of two-dimensional patern on nonlinear Coordinates is necessary for visual feedback. We propose a resistive network for this purpose and report an analysis of its errors.

This nonlinear mapping function can be used in other applications. For example, it can be used for foveated vision to extract the centroid.