Normal Optical Flow Imager
Various motion properties of an image can be computed by using Normal
Flow measurements. Notable among these are focus of expansion, time to
contact, etc. VLSI systems have been developed which combine processing
with imaging but almost always the imaging resolution suffers in such
cases.
We designed a two-dimensional dense Normal Optical Flow measurement
chip implemented in 0.5 micron CMOS process which combines imaging and
processing on the same chip efficiently. The algorithm computes partial
derivatives with respect to time and space and uses their ratio to
compute normal flow velocity.
The chip has a 92x52 array of APS pixels, occupies an area of 4.5 mm2
and consumes 2.6mW power. It has two CDS circuits to improve the noise
characteristics of the present and previous frame image. We use V-to-I
circuits to compute spatial and temporal gradients and ratio circuits
to compute the velocities. We also have confidence measures available
to remove unreliable results where the spatial derivatives are below a
certain threshold, thus making the ratios singular.