Color can be a useful feature in autonomous vehicle systems that are based on machine vision, for tasks such as obstacle detection, lane/road following, and recognition of miscellaneous scene objects. Unfortunately, few existing autonomous vehicle systems use color to its full extent, largely because color-based recognition in outdoor scenes is complicated, and existing color ma-chine vision techniques have not been shown to be e ective in realistic outdoor images. This paper presents a technique for achieving e ective real-time color recognition in outdoor scenes. The technique uses Multivariate Decision Trees for piecewise linear non-parametric func-tion approximation to learn the color of a target object from training samples, and then detects targets by classifying pixels based on the approximated function. The method has been success-fully tested in several domains, such as autonomous highway navigation, o-road navigation and target detection for unmanned military vehicles, in projects such as the U.S. National Automated Highway System (AHS) and the U.S. Defense Advanced Project Agency- Unmanned Ground Vehicle (DARPA-UGV). MDT-based systems have been used in stand-alone mode, as well as in conjunction with systems based on other sensor con gurations.

Statistics-based colour constancy algorithms work well as long as there are many colours in a scene, they fail however when the encountering scenes comprise few surfaces. In contrast, physics-based algorithms, based on an understanding of physical processes such as highlights and interreflections, are theoretically able to solve for colour constancy even when there are as few as two surfaces in a scene. Unfortunately, physics-based theories rarely work outside the lab. In this paper we show that a combination of physical and statistical knowledge leads to a surprisingly simple and powerful colour constancy algorithm, one that also works well for images of natural scenes.

This paper analyzes the variation of the color of objects with respect to existing models of daylight and surface reflectance, and shows that the apparent color of objects can be represented by characteristic distributions in RGB space. These distributions can "learned" from training samples using non-parametric approximation techniques. Two approximation techniques -- Multivariate Decision Trees (MDT's) [4] and Neural Networks (with Back-propagation) [36] are used to approximate decision boundaries around the training samples. Image pixels are then classified according to their location with respect to the learned decision boundary. The results from the two techniques are very similar, suggesting that the choice of the approximation/ clas2 sification technique is not very crucial, as long as the technique is capable of approximating arbitrary RGB distributions given adequate training samples. The MDT-based classification technique has been tested in a number of domains, such as autonomous highway navigation, off-road navigation, military target detection and skin finding; results from some of these tests are described. 2 Causes for color shift in outdoor scenes

This paper analyzes the variation of the color of objects with respect to existing models of daylight and surface reflectance, and develops context-based models of daylight (based on the CIE model [18]) and hybrid surface reflectance (based on existing hybrid surface reflectance models [21, 25, 32, 35]) called the Normalized Photometric Function. Thereafter, given the time-of-day (which, along with location, is used to calculate the sun-angle [23]), approximate cloud cover and sun-visibility, the color of the incident daylight is predicted, and combined with the reflectance model of the target object to predict the apparent color of the object; image pixels are then classified based on the predicted color. Section 2 describes the causes for the variation in apparent color; section 3 gives a brief literature review; section 4 describes the CIE daylight model and the context-based daylight model developed in this study; section 5 analyzes surface reflectance with respect to existing models and then develops the Normalized Photometric Function (NPF) model; section 6 combines the daylight and NPF models for context-based color prediction; finally, section 7 summarizes the conclusions of the study. 2 Causes for color shift in outdoor scenes

The present work demonstrates for the first time the advantage of an analytical Gaussian model of spectral function approximation for solving the colour constancy problem. This model was compared with well-known linear models in numerical simulations performed over an extensive set of natural pigments illuminated by both ideal Planck and real light sources. The precision and stability of the colour constancy were estimated using the same colour constancy algorithm for all the models compared. The experiments indicate that the Gaussian model is potentially more effective for estimating the chromaticity of the scene illuminant and the scene objects than the linear spectral models are. In addition, the Gaussian model involves the possibility of comparing the object colour with the etalon colour captured by another sensor.

Realistic image synthesis provides principles and techniques for creating realistic imagery based on models of real-world objects and behaviors. It has widespread applications in 3D design, computer animation, and scientific visualization. While it is common to describe light and objects in terms of colors, this approach is not sufficiently accurate and cannot render many spectral phenomena such as interference and diffraction. Many researchers have explored spectral rendering and proposed several methods for spectral representation, but none satisfy all representation criteria such as accuracy, compactness and efficiency. Furthermore, previous studies have focused on distinct behaviors of natural phenomena but few on their commonality and generality, and it is difficult to combine existing algorithms to simulate complex processes. This thesis proposes solutions to these problems within a spectrum-based rendering framework. The pipeline begins by loading spectra from a database to specify light sources and objects, then generates a spectral image based on local and global illumination models, projects the spectral image into a CIE image, and finally converts the CIE image into an RGB image for display or a CIELab image for evaluation. In spite of omitting the light phase information, it is