262 
4. EXPERIMENT RESULTS 
4.1 Test pixel 
A one-pixel experiment chip is designed and fabricated by using the Mietec technology, 2.4um (double-metal double- 
poly CMOS). The complet schematic of this pixel is depicted in Figure 2. The layout occupies about 247x1 95um^?, but is 
not optimazed. The effects of the adaptation can be observed on the figure 5. The threshold voltage, memorized in the 
analog memory, follows the lighting evolution. 
The figure 6 has been obtained by moving the sensor with a sine trembling movement behind a lighting spatial edge. The 
photodiode voltage at the end of each exposure time has a sine form. This figure is an example of the translation of a 
spatial information, which is visualised by the "trembling movement" curve, into a temporal information, which is 
visualised by Vphoto(texp). 
4.2 Analog smart image sensor 
The figure 7 shows a local edge image extraction. A simple geometrical object is moving over the sensor with a trembling 
movement. The trembling movement direction is perpendicular to the visible sides of the object. The obtained image has 
been recuperated after the simple processing described earlier. Gray zones are adapted pixels zones, the white one 
corresponds to a positive lighting variation, and the black one to a negative variation. The black point in the middle of the 
object corresponds to a defective pixel. We can notice that we have no information on the nature of the contrast 
between the object and the bottom (which one is the most dark ?) since each pixel has adjusted its sensitivity threshold 
to its local light level. We note that the parallel side is invisible since the pixels situated on this edge can not detect 
temporal lighting variation. With two perpendicular axes movement, the sensor would be able to extract the complete 
outlines of this object. On the first image, the left side line is black and the right one is white. This is comprehensible 
because for one direction of the vibration, all pixels of one side detect a lighting diminution, and all others detect an 
increase. The opposite image is obtained for the other direction of vibration. 
5. CONCLUSIONS 
This paper presents an analog adaptive smart image sensor. The main advantages of this sensor is its ability to 
translate spatial informations into temporal ones and its ability to eliminate the effects of local ambiant light on detection 
by adaptation process. Likewise, this permit to reduce FPN problems. The local edge image extraction presented here 
shows the sensor's possibilities with modest microprocessor interface. We plan to interface this adaptive sensor with an 
integrated massively parallel processing array for some dedicated low cost vision tasks. 
  
        
    
  
NE. 
7N 
5 I I 1 T | I 1 1 T | T T I I T T T T T I T T 
——3X—— Light level 7 
4 c9 Vo Nr 
3 = 
S 
= 2 T] 
o j 
> = 
With light Without light 
  
  
© — 
DIT Eg FTYTTITTTTIT[TUTTTTT" 
X 
X 
X 
LL | | 
  
  
  
© 
A 
no 
C 
A 
O1 
Time (s) 
Figure 5: Effect adaptation on the capacitor voltage. 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995