Correlation Image Sensor for Algebraic Solution of Optical Flow
Shigeru Ando, Masanori Nagase, Takashi Watanabe, Tomohiko Kosugi, Tetsuya Iida
Abstract
In this article, we describe several enhancements of a three-phase correlation image sensor (3PCIS) toward its uses for a direct algebraic method of optical flow detection, i.e., determining velocity field of image patterns for detecting, tracking, and three-dimensional recovery of objects in a scene. A recent theoretical study has provided an exact closed-form solution of the velocity from temporal Fourier integral measurements of a time-varying intensity distribution. An accurate analog multiplication and integration of incident light and reference signals required for this type of measurements is the inherent advantage of a 3PCIS. We fabricate and compare several types of correlation detection block. As continuous-value inputs of the algebraic solution, compensation operations of systematic sources of error are so designed to realize a near shot-noise-limited performance. For random telegraph noise and interference reduction, we show that a p-epi channel, deep p-well structure is the best 3P multiplier. Several experimental results of optical flow detection including an accuracy evaluation and the dense motion capture of a traffic scene are given.