Physically customized DEFORMOTION models for electro-optic sensor data

ABSTRACT: In the field of computer vision, extraction of features from sensor data (EO/IR) is critical for inference tasks including target object detection, recognition, classification, segmentation and tracking. The objective of this proposal is to propose computation deformotion models that incorporate various physical properties of scene objects, particularly passenger vehicles. Layered deformotion algorithms are model-based image analysis algorithms that account for both geometric (shape and motion) as well as radiance features of scene objects. These models however, do not account for nuisance factors in radiance models, which cause mismatch between object appearance. This proposal aims to incorporate mathematical heat diffusion models for materials typically used in vehicles, in the radiance features of deformotion models. This proposal further aims to investigate other physical properties of passenger vehicles that can be incorporated into the layered deformotion models to model and mitigate the radiance model related nuisance factors, leading to more robust performance of geometric shape and appearance matching algorithms. BENEFIT: The proposed research will provide concrete mathematical models for incorporating mathematical models based on physical properties of passenger vehicles. These models will enable testing on real-world data in phase - II and eventually development of software applications useful for recognition related tasks. These applications could then be deployed for perimeter security of important buildings and installations, in Unmanned Grround and Aerial Vehicles and would be useful in both civil and defense applications.