32×32 winner-take-all matrix with single winner selection

et al. 2019

Sensors

Winner-Take-All (WTA) circuits play an important role in applications where a single element must be selected according to its relevance. They have been successfully applied in neural networks and vision sensors. These applications usually require a large number of inputs for the WTA circuit, especially for vision applications where thousands to millions of pixels may compete to be selected. WTA circuits usually exhibit poor response-time scaling with the number of competitors, and most of the current WTA implementations are designed to work with less than 100 inputs. Another problem related to the large number of inputs is the difficulty to select just one winner, since many competitors may have differences below the WTA resolution. In this paper, a WTA circuit is presented that handles more than four thousand inputs, to our best knowledge the hitherto largest WTA, with response times below the microsecond, and with a guaranty of just a single winner selection. This performance is obtained by the combination of a standard analog WTA circuit and a fast digital single-winner selector with almost no size penalty. This WTA circuit has been successfully employed in the fabrication of a Selective Change-Driven Vision Sensor based on 180 nm CMOS technology. Both simulated and experimental results are presented in the paper, showing that a single pixel event can be selected in just 560 ns, and a multipixel pixel event can be processed in 100 μs. Similar results with a conventional approach would require a camera working at more than 1 Mfps for the single-pixel event detection, and 10 kfps for the whole multipixel event to be processed.

Section: State Of the Artmentioning

confidence: 99%

A 4K-Input High-Speed Winner-Take-All (WTA) Circuit with Single-Winner Selection for Change-Driven Vision Sensors

Pardo

Reig

et al. 2019

Sensors

“…The first difference in the SCD sensor is that each pixel has its own analogue memory with the last read-out value. The difference between the current and the stored value is compared in parallel for all pixels in the sensor using a Winner-Take-All circuit [ 16 ]. The pixel with the greatest change is selected first, and its illumination level and coordinates are read out for processing, the local stored value is updated, so this pixel will lose all next voting until there is, again, a large difference in its illumination level.…”

Section: Selective Change-driven Systemmentioning

confidence: 99%

Taking Advantage of Selective Change Driven Processing for 3D Scanning

Vegara

Zuccarello

et al. 2013

Sensors

This article deals with the application of the principles of SCD (Selective Change Driven) vision to 3D laser scanning. Two experimental sets have been implemented: one with a classical CMOS (Complementary Metal-Oxide Semiconductor) sensor, and the other one with a recently developed CMOS SCD sensor for comparative purposes, both using the technique known as Active Triangulation. An SCD sensor only delivers the pixels that have changed most, ordered by the magnitude of their change since their last readout. The 3D scanning method is based on the systematic search through the entire image to detect pixels that exceed a certain threshold, showing the SCD approach to be ideal for this application. Several experiments for both capturing strategies have been performed to try to find the limitations in high speed acquisition/processing. The classical approach is limited by the sequential array acquisition, as predicted by the Nyquist–Shannon sampling theorem, and this has been experimentally demonstrated in the case of a rotating helix. These limitations are overcome by the SCD 3D scanning prototype achieving a significantly higher performance. The aim of this article is to compare both capturing strategies in terms of performance in the time and frequency domains, so they share all the static characteristics including resolution, 3D scanning method, etc., thus yielding the same 3D reconstruction in static scenes.

“…Every pixel has an analogue memory with the last read-out value. The absolute difference between the current and the stored value is compared for all pixels in the sensor; the pixel that differs most is selected using a Winner-Take-All circuit (WTA) (Zuccarello et al, 2010) and its illumination level and coordinates are read out for processing. It is important to note that the concept of snapshot at instant t can be kept, taking into account that all the photodiode charges begin and finish at the same time.…”

Section: Scd System and Processingmentioning

confidence: 99%

On the Design of Change-driven Data-flow Algorithms and Architectures for High-speed Motion Analysis

Proceedings of the 9th International Conference on Informatics in Control, Automation and Robotics

Zuccarello

Pardo

et al. 2012

Motion analysis is a computationally demanding task due to the large amount of data involved as well as the complexity of the implicated algorithms. In this position paper we present some ideas about data-flow architectures for processing visual information. Selective Change Driven (SCD) is based on a CMOS sensor which delivers, ordered by the absolute magnitude of its change, only the pixels that have changed after the last time they were read-out. As a natural step, a processing architecture based on processing pixels in a dataflow method, instead of processing complete frames, is presented. A data-flow FPGA-based architecture is appointed in developing such concepts.