An analog VLSI implementation of real-time motion detection in early vision using a hybrid of the Reichardt and Ullman-Marr models has been fabricated. It has a 5×5 array of photoreceptors, with a receptor fill factor of 0.42. The majority of this circuit operates in strong inversion. The photosensors have squared logarithmic and linear responses for low and high intensity illumination, respectively. A linear passive resistive grid is used to implement the desired difference of Gaussian impulse response transfer function of the edge detection circuit whose dynamic range is 3 V. These outputs are used to determine the velocities of the translating objects. Velocities up to 2 m/s on chip corresponding to 6 m/s in space have been measured in 2D. The array has been fabricated in a 2 μm n-well CMOS process
Published in:
Neural Networks, 1992. IJCNN., International Joint Conference on
(Volume:4
)
Date of Conference: 7-11 Jun 1992