[4],
her
on.
the
ed,
(n-
ght
ain
an
25
297
In the conditions mentioned above, a first prototype circuit (128 pixels) has been designed and submitted to
fabrication in a 2.4um CMOS standard process. First hand experience shows that pixel dimensions can be reduced to
16umx16pm in a 1um CMOS technology. Because of the high speed (5 MHz non-overlapping clock) of the Gaussian
filter, tested on the previous chip [4], we expect a processing rate of 2.56ms for an 256x256 image. To our knowledge,
this datum gives the best stereofusion throughput at good illumination conditions. This gives us the possibility to
envision a temporal averaging process without loosing real-time aspects.
The 2D prototype have to realize a linear conversion of the light-generated pixel charge in the active matrix area
into voltage in the read-out stage [5]. The main problem with this is the parasitic capacitance of each column bus,
which is often bigger than the pixel one. A charge amplifier module between the photodiode matrix and the Gaussian
filter is needed on each chip in order to minimize those effects. A second prototype circuit ( 64x64 pixel matrix) has
been designed and submitted to fabrication in a {um CMOS standard process.
5. CONCLUSIONS
This paper presents an obstacle-detection tailored stereovision algorithm and its associated analog VLSI
implementation architecture. Our investigation shows the possibility to realize fast, dense and low-cost CMOS
monolithic stereovision sensors for the navigation-based applications.
ACKNOWLEDGEMENTS
This work is supported by GIE P.S.A. Peugeot Citroen.
REFERENCES
[1] Marr D., T. Poggio, 1979 "A Theory of Human Stereo Vision" Proc. Roy. Soc.London, Vol.B204
[2] Enkelmann & al., 1988 "An Experimental Investigation of Estimation Approaches for Optical Flow Fields"
in Motion Understanding - Robot and Human Vision, edited by W.N. Martin , J.K. Aggarwal,
Kluwer Academic Press
[3] Marr D., E. Hildreth, 1980 "Theory of Edge Detection" Proc. Hoy. Soc. London, Vol.B207
[4] Ni, Y., B. Arion, F. Devos, 1993 "Yet Another Gaussian Convolver" Proc. 1993 IEEE International
Symposium on Circuits and Systems, ISCAS'93 Chicago
[5] Jansson, C., P. Ingelhag, C. Svenson, R. Forchheimer, 1993 "An Adressable 256x256 Photodiode Image
Sensor Array with an 8-bit Digital Output" , Analog Integrated Circuits and Signal Processing,
Vol4, No 1, pp 37-49
IAPRS, Vol.-30, Part 5W1, ISPRS Intercommision Workshop "From Pixels to Sequences", Zurich, March 22-24, 1995