Single-Chip High-Speed Computation of Optical Flow

In the second part of the article we demonstrate how the VIP-chip is able to perform the computation at This Paper consists of two Parts. In the first Part near-video rate (11-15 frames/second). ~h~ m c h i p we describe how to compute optical flow from second is by a control unit and a limited set of derivatives. In the second part we describe the VIP external memories. chip and how this chip can compute a 512x512 optical flow field at nearly full video rate. OPTICAL FLOW FROM SECOND DERIVATIVES The basic equation for optical flow is given by (1). INTRODUCTION fx.u + f , . v + f , = 0 (1) Optical flow is a means for the analysis of image sequences. In particular it should be helpful in segmenting rigidly moving objects from a non-moving background or from a background the movement of which could be predicted. The latter case is very often at hand when the image sequence comes from a moving camera in a 3D-world. The potential applications of optical flow and various segmentation algorithms based thereof are to be found in target tracking, automatic inspection, computer vision for autonomous robots and vehicles and related areas. These applications have been slow in coming because of computation complexity and high cost. In this paper we will address this problem in two ways. Firstly, we advocate the use of second derivatives to solve the optical flow equation explicitly. Secondly, we will present a new chip, called WP (VIdeo Processor), which comprises 512 bit-serial processors and seems rather ideal for the task. In the first part of the article we demonstrate that one can obtain a useful optical flow field from the second derivatives using the following computations. 25 to 50 operations of type addition or subtraction, the number depending on noise levels which require more or less smoothing in the operators.