Analog computation provides an efficient parallel architecture that can be realized in smaller silicon areas with reduced power consumption and higher processing speed, as compared to digital implementation in areas such as vision processing and neural network implementations. This research paper will be of interest to people with a background in analog signal processing and computation, and its very large-scale integration (VSLI) implementation.
Stocker presents an implementation of a focal plane VLSI sensor that estimates optical flow in two dimensions. This implementation is based on a model that provides a locally smooth optical flow with a controllable degree of smoothness. The model of the optical flow sensor consists of a two-layer network of locally connected arrays of computational units. Each optical flow unit in the array acts in a feedback loop, comparing its local visual information with the collective information of the optical flow field.
The main contributions of the paper are: a new model that provides an optimization problem for the two-dimensional optical flow by providing spatial integration of visual information; an implementation of an array of analog computational optical flow units in a bipolar complementary metal oxide semiconductor (BiCMOS) process; logarithmic photoreceptor circuits that make the sensor independent of absolute illumination in the visual scene; a good list of references; and a review of prior work done in the area of optical flow sensors. One aspect that is not properly addressed in the paper is the effect of mismatches of devices on a system’s overall performance.