We report on a vision sensor architecture relying on a pixel-embedded memristive device to perform dynamic background subtraction as basic image processing components, aimed at detecting anomalous events in the scene. A Light-To-Frequency Converter generates digital pulses, which are linearly proportional with the light intensity. During the exposure time, a certain number of pulses, with proper width and amplitude, drive the memristor MS, changing its resistance. The value of MS is then compared with two memristors, MH and ML, used as thresholds. After each frame, the values of MH and ML are adjusted in order to satisfy ML < MS < MH. The rate at which the two memristors are updated is application-dependent and can be digitally programmed. If the value of MS trespasses either MH or ML, due to a light change, the pixel behaviour is recognized as anomalous (hot-pixel) By aggregating the hot-pixels of the image, anomalous events can be detected through standard processing techniques. The pixel architecture has been designed in a 0.35µm standard CMOS process and validated through simulation.

Memristor-based pixel for event-detection vision sensor

Olumodeji, Olufemi Akindele;Gottardi, Massimo
2015

Abstract

We report on a vision sensor architecture relying on a pixel-embedded memristive device to perform dynamic background subtraction as basic image processing components, aimed at detecting anomalous events in the scene. A Light-To-Frequency Converter generates digital pulses, which are linearly proportional with the light intensity. During the exposure time, a certain number of pulses, with proper width and amplitude, drive the memristor MS, changing its resistance. The value of MS is then compared with two memristors, MH and ML, used as thresholds. After each frame, the values of MH and ML are adjusted in order to satisfy ML < MS < MH. The rate at which the two memristors are updated is application-dependent and can be digitally programmed. If the value of MS trespasses either MH or ML, due to a light change, the pixel behaviour is recognized as anomalous (hot-pixel) By aggregating the hot-pixels of the image, anomalous events can be detected through standard processing techniques. The pixel architecture has been designed in a 0.35µm standard CMOS process and validated through simulation.
978-1-4799-8203-5
978-1-4799-8203-5
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11582/302189
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
social impact