Image accuracy and representational enhancement through low-level, multi-sensor integration techniques

Multi-Sensor Integration (MSI) is the combining of data and information from more than one source in order to generate a more reliable and consistent representation of the environment. The need for MSI derives largely from basic ambiguities inherent in our current sensor imaging technologies. These ambiguities exist as long as the mapping from reality to image is not 1-to-1. That is, if different 44 realities'' lead to identical images, a single image cannot reveal the particular reality which was the truth. MSI techniques can be divided into three categories based on the relative information content of the original images with that of the desired representation: (1) ''detail enhancement,'' wherein the relative information content of the original images is less rich than the desired representation; (2) ''data enhancement,'' wherein the MSI techniques axe concerned with improving the accuracy of the data rather than either increasing or decreasing the level of detail; and (3) ''conceptual enhancement,'' wherein the image contains more detail than is desired, making it difficult to easily recognize objects of interest. In conceptual enhancement one must group pixels corresponding to the same conceptual object and thereby reduce the level of extraneous detail. This research focuses on data and conceptual enhancement algorithms. To be useful in many real-world applications, e.g., autonomous or teleoperated robotics, real-time feedback is critical. But, many MSI/image processing algorithms require significant processing time. This is especially true of feature extraction, object isolation, and object recognition algorithms due to their typical reliance on global or large neighborhood information. This research attempts to exploit the speed currently available in state-of-the-art digitizers and highly parallel processing systems by developing MSI algorithms based on pixel rather than global-level features