A retina-inspired robust on-focal-plane multi-band edge-detection scheme for CMOS image sensors

Abstract

An edge-detection scheme suitable for machine vision and digital motion detection applications is presented. The scheme is inspired by the human visual system (human retina) and modified for a compact and scalable CMOS hardware implementation. In addition, pixel circuit and implementation of the scheme on a CMOS focal-plane are proposed and simulation results are presented. The circuit performs thresholding and single-bit quantization of high-frequency image content over multiple frequency bands using a low-complexity down-sampling scheme. As a result, a 1-bit per pixel output is obtained which results in efficient edge-detection of the image content. Depending on the number of the utilized frequency bands, the presented scheme achieves image compression levels that range from 1 BPP (bit-per-pixel) to 1.33 BPP. The proposed acquisition technique and the corresponding CMOS circuit can easily be adjusted to various imaging applications and scaled towards new CMOS technology nodes and high resolution image sensors.

Distributed mixed-signal architecture for programmable smart image sensors

“…During the last decade several smart vision sensors have been designed in standard CIS technology [1,[3][4][5][6][7][8][9][10][11][12][13][14][15][16][17]. The increasing resolutions and frame rates result in a large data transfer between the imaging array and the processing unit.…”

“…Since our work is aimed at tightly-embedded applications, with strong requirements in terms of power consumption and surface, analog implementations seem more appropriate. Therefore, we focus here on analog implementations of common processing tasks such as edge detection using spatial convolution [1,5], difference of averaged images [6] and neighbours comparison [7] ; motion detection using temporal difference [1,7] ; or image enhancement [1,8].…”

Distributed mixed-signal architecture for programmable smart image sensors

“…During the last decade several smart vision sensors have been designed in standard CIS technology [1,[3][4][5][6][7][8][9][10][11][12][13][14][15]. The increasing resolutions and frame rates result in a large data transfer between the imaging array and the processing unit.…”

“…On the other hand, analog processing needs no in-pixel ADC, and would consume less power. Therefore, and considering our application, we focus here on analog implementation of common processing tasks such as edge detection using spatial convolution [1,3], difference of averaged images [4] and neighbours comparison [5] ; motion detection using temporal difference [1,5] ; or image enhancement [1,6].…”

“…Virtual macropixels are used for region-of-interest detection: pixels are processed together as a virtual cluster by out-of-the-matrix electronics or software. For example, this method is applied for spatial averaging [4,10], computing of local integration time [11] or memory optimization by pixel interlacing [12]. On the other hand, the concept of macropixels can be implemented in hardware by mutualizing in-matrix circuitry for the block of pixels instead of repeating it in every pixel.…”