An analog VLSI velocity sensor using the gradient method

Smart vision sensors that unify imaging and computation on one single chip offer great advantage over conventional sensor systems, where the computational part is usually performed separately and serially on a digital computer. Using parallel analog VLSI design principles, compact, low power, inexpensive and real time sensors can be built even in standard CMOS processes. In this paper we present the first working analog VLSI implementation of a 1-D velocity sensor that uses the gradient method for spatially resolved velocity computation. We use a novel floating gate wide linear range amplifier and a floating gate division circuit. The division by zero problem of the gradient method has been solved. The sensor velocity output linearly codes the stimulus velocity over a wide range, is independent of contrast down to 20% contrast and indicates the correct direction-of-motion down to 4% contrast. In a circular pixel arrangement the sensor reports the rotational velocity up to 350 rpm.