Robust Integration of Thermal and Visual Imagery for OutdoorScene Analysis

点击翻译按钮获取中文摘要

Past research by this investigator established a new method for analyzing outdoor scenes. Thermal and visual imagery was combined to extract internal object properties that serve as physically meaningful features for region labeling. This new research presents an improved formulation of the energy exchange model used for this purpose. The energy exchange model is formulated as a linear regression model. Statistically robust techniques will be used to extract reliable estimates of internal object properties such as thermal capacitance. These techniques allow reliable estimates in spite of almost 50% of the data being arbitrarily corrupted due to misregistration of images and segmentation errors. Statistically robust techniques have been shown to be beneficial in many important computer vision problems such as pose estimation, straight-line extraction, computation of image structure, and surface fitting. The proposed research will explore statistical robust approaches to another important problem, i.e., sensor fusion. The computational characteristics of the robust algorithms will be explored for the chosen sensor fusion task. The new formulation also allows iterative refinement of surface parameters (such as thermal emissivity) which were assumed known in the previous formulation of the model. The new approach thus improves the accuracy of the internal property value estimates and also provides new information regarding surface parameter values. The method thus provides a greater number of stable features for object classification.