Using the Image Understanding Architecture and Knowledge Based Reasoning to Recognized Vehicle Types from Multisensor Data

Current Army projects, the Intelligent Vehicle Highway System, and the Unmanned Ground Vehicle project require automated signal processing and information processing techniques that distinguish various types of vehicles in real-time. Tasks, such as distinguishing 3-axle trucks from 2-axle trucks or trucks from cars, require image understnding, artificial intelligence, and high performance computing. We proposed to use the advanced processing architecture of the Image Understanding Architecture (IUA) in an AI paradigm that combines these various forms or processing. In particular, we proposed to utilize the IUA for recognizing vehicle types from multisensor data fusion, and then using sensor and domain knowledge to control the extraction, grouping, and matching of image features against stored abstract models fo the various vehicle types. In Phase I, we will use static images of military targets to experiment with extracting various image features such as lines, arcs, and regions from both visible and IR images. These features will be perceptually organized to form sysmbolic descriptions of complex imaghe events. Model matching algorithms will be developed to classify the image events based on abstract models of known vehicles. This approach will be tested by measuring the accuracy and speed of the model matching process on the IUA.