(A) study on direct visual servoing with field-rate COTS vision system

Abstract

In this dissertation, we concentrate on development of quantitative models of measurement noise and time-delay in field-rate COTS vision system. And also we show their value by an application of direct visual servoing. We analyze operations of the vision systems physically and electrically and then derive mathematical models by using adequate methods of some limitations and simplifications. At last, we verify their validity by experiments. Although it is desired that the models may be proved for all features, we consider only posture features such as position and angle in this dissertation. For the other features, we can derive models by using similar methods as mentioned in the dissertation. Chapter 2 gives basic backgrounds needed for developments in the following chapters. First, a definition of field-rate COTS vision system is given. Next, analysis in frequency domain of optical lens system of camera is given and followed by color detection algorithm and its fast version, which is used in experiments. And mathematical definitions and error characteristics of geometric features used in the dissertation are given. Last, statistical verification methods such as Kolmogorov-Smirnov test and W test are described, which are widely used in problems of deviation between an empirical and a theoretical distribution or between unknown distributions. In Chapter 3, a noise model of postures measured by off-the-self vision systems is proposed by using a probabilistic binary image model. Due to deterministic and stochastic effects of off-the-shelf vision systems, we get noisy bitmap images with intermediate colors on their boundaries. After color detection, instead of a pure binary image representation, we use the probabilistic binary image model, which consists of interior region, exterior region, and boundary region: Binary values of pixels in the boundary region are defined as i.i.d. binary random variables with a constant probability called modal probability constant....