TY - JOUR
T1 - Analog VLSI implementation of spatio-temporal frequency tuned visual motion algorithms
AU - Higgins, Charles M.
AU - Pant, Vivek
AU - Deutschmann, Rainer
N1 - Funding Information:
Manuscript received January 12, 2004; revised June 18, 2004. The work of C. M. Higgins and V. Pant was supported by the Office of Naval Research under Grant N68936-00-2-0002. The work on the Barlow–Levick sensor was supported by the Office of Naval Research. This paper was recommended by Associate Editor G. Cauwenberghs.
PY - 2005/3
Y1 - 2005/3
N2 - The computation of local visual motion can be accomplished very efficiently in the focal plane with custom very large-scale integration (VLSI) hardware. Algorithms based on measurement of the spatial and temporal frequency content of the visual motion signal, since they incorporate no thresholding operation, allow highly sensitive responses to low contrast and low-speed visual motion stimuli. We describe analog VLSI implementations of the three most prominent spatio-temporal frequency-based visual motion algorithms, present characterizations of their performance, and compare the advantages of each on an equal basis. This comparison highlights important issues in the design of analog VLSI sensors, including the effects of circuit design on power consumption, the tradeoffs of subthreshold versus above-threshold MOSFET biasing, and methods of layout for focal plane vision processing arrays. The presented sensors are capable of distinguishing the direction of motion of visual stimuli to less than 5% contrast, while consuming as little as 1 μW of electrical power. These visual motion sensors are useful in embedded applications where minimum power consumption, size, and weight are crucial.
AB - The computation of local visual motion can be accomplished very efficiently in the focal plane with custom very large-scale integration (VLSI) hardware. Algorithms based on measurement of the spatial and temporal frequency content of the visual motion signal, since they incorporate no thresholding operation, allow highly sensitive responses to low contrast and low-speed visual motion stimuli. We describe analog VLSI implementations of the three most prominent spatio-temporal frequency-based visual motion algorithms, present characterizations of their performance, and compare the advantages of each on an equal basis. This comparison highlights important issues in the design of analog VLSI sensors, including the effects of circuit design on power consumption, the tradeoffs of subthreshold versus above-threshold MOSFET biasing, and methods of layout for focal plane vision processing arrays. The presented sensors are capable of distinguishing the direction of motion of visual stimuli to less than 5% contrast, while consuming as little as 1 μW of electrical power. These visual motion sensors are useful in embedded applications where minimum power consumption, size, and weight are crucial.
KW - Analog very large-scale integration (VLSI)
KW - Biomimetic
KW - Spatio-temporal frequency
KW - Vision chip
UR - http://www.scopus.com/inward/record.url?scp=16344391425&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=16344391425&partnerID=8YFLogxK
U2 - 10.1109/TCSI.2004.841601
DO - 10.1109/TCSI.2004.841601
M3 - Article
AN - SCOPUS:16344391425
SN - 1057-7122
VL - 52
SP - 489
EP - 502
JO - IEEE Transactions on Circuits and Systems I: Regular Papers
JF - IEEE Transactions on Circuits and Systems I: Regular Papers
IS - 3
ER -