wn
CD
Cif
ure
to
era
34.
1ge
259
AN ANALOG ADAPTIVE SMART IMAGE SENSOR FOR SPATIO-TEMPORAL
INFORMATION EXTRACTION
F. Lavainne*, Y. NI**, P. de Carné*, F. Devos*
* Institut d'Electronique Fondamentale, Université Paris Sud
91405 Orsay Cedex, France
Phone: (33) 1 69 41 78 32; Email: lavainne @axis.ief-paris-sud.fr
** Institut National des Télécommunications
9, rue Charles Fourier
91011 Evry Cedex, France
Email: ni@galaxie.int-evry.fr
KEYWORDS: adaptive image sensor, FPN, low level vision, spatio-temporal informations, local edge image.
ABSTRACT
This paper presents an analog adaptive image sensor, which can extract spatio-temporal salient informations, without
fixed pattern noise (FPN) problems. Some low level vision problems can be done with only simple local processing.
Each pixel of this smart sensor has a photoreceptor (PN junction) and a tiny analog processor. The tiny analog
processor adjusts constantly by small steps the local sensitivity according to the average incoming lighting level. This
adaptation process makes the sensor insensible to the slow ambiant light change but very sensitive to fast variations.
The digital output of this smart sensor can be readout fastly of the sensor without noise-speed problems such as in
analog imager sensors. Many salient spatio-temporal informations can be extracted by analysing these digital output
signals. In the natural image sensing conditions, spatial contrast informations can be translated into pure local temporal
one by trembling movement of the sensor. By sensing these temporal informations, spatial contrast informations, such
as local edge image, can be extracted without the FPN problem in the neighbour pixels because the spatial informations
are sensed by the same pixel associated with sensor trembling movement.
1. INTRODUCTION
Many obstacles exist for the construction of artificial visual systems as like human eye. A practical method is to use
some structured light sources so as to simplify vision tasks (Besl,1988). Two structured light types can be used: the
spatial structured light and the temporal structured light; they can be combined. An interest for the biological
environment shows that the human visual system is mainly sensitive to the temporal component : an image that remains
strictly static in front of the retina becomes rapidly invisible; this is due to a biological adaptation process (Yarbus,
1967). In the natural environment, the temporal structured light is therefore the most interesting.
Some smart sensors have been studied, but these circuits either use only spatial structured light (Gruss,1991),
(Ni,1993) either use temporal structured light (Delbrück, 1994) but with an adaptation constant time fixed and not
programmable. In this paper, we present an adaptive image sensor for the temporal structured light. Each pixel of this
sensor has a photoreceptor and a tiny analog processor. This tiny analog processor adjusts constantly the local
sensitivity according to the ambiant light level. It is thus insensitive to the slow variations, but on the other hand very
sensitive to quick variations, such acts a high pass filter. The speed of adaptation process is programmable. This
increases the dynamic range of detection. It is thus possible, by making trembling movement to the sensor, to translate
spatial contrast informations into pure local temporal ones. By sensing these temporal informations, spatial contrast
informations can be extracted.
2. PIXEL STRUCTURE
2.1 Photoreceptor
There are principaly two types of photoreceptors: discrete photoreceptor and continuous one. The discrete one uses
some photocharge integration such as that in CCD or in MOS imager. The continuous one uses somewhat current-
voltage conversion structure (Mead, 1989). The advantage of the discrete configuration is the conservation of short
light variations, contrary to the continuous configuration (Lavainne, 1994). This can give a better sensitivity in the case
where perfect synchronization is not possible. : iin
We have chosen the discrete photoreceptor, a diffused diode on the substrate, because in many applications a good
synchronization is not possible. The discrete aspect is obtained by a reverse polarisation of the diode, followed by an
insulation of it from polarisation during a fixed time.
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995