Geometric, Algebraic, and Thermophysical Techniques for Object Recognition in IR Imagery

摘要

We describe a new approach for computing invariant features in infrared (IR) images. Our approach is unique in the field since it considers not just surface reflection and surface geometry in the specification of invariant features, but it also takes into account internal object composition and thermal state which affect images sensed in the nonvisible spectrum. We first establish a nonlinear energy balance equation using the principle of conservation of energy at the surface of the imaged object. We then derive features that depend only on material parameters of the object and the sensed radiosity. These features are independent of the scene conditions and the scene-to-scene transformation of the ``driving conditions'' such as ambient temperature and wind speed. The algorithm for deriving the invariant features is based on the algebraic elimination of the transformation from the nonlinear object-image relationships. The elimination approach to compute absolute invariants is a general method based on deriving and separating the polynomial relations which are invariant with respect to the given transformation. A complete model-based approach for recognition of objects in IR images is presented. Geometric invariant features are used to generate hypotheses of object identity and pose. These hypotheses are verified or refuted by the thermophysical features. Results on real IR imagery are shown to illustrate the performance of the features and the methodology in an object recognition system that deals with multiple classes of objects.