The Current Real Time (RTV) Systems are known to have many Limitations. They are not robust in many real life scenarios which include target occlusion, non ideal environmental conditions and counter measures. Due to the difficulties that we have in dealing with the uncertainties originating in the processors, sensors, and atmospheric turbulence, the occlusion problem has been ignored by the researches and developers of the RTVs to a large extent. However, for the case of relatively small number of objects such as those encountered in tactical military environment the problem seems to have a conceptually simpler solution than the general object recognition problem. To enable robust real time extraction of object information from image sequence we propose a technique similar to those used for video and image compression purposes - a wavelet transform similar, which allow for the removal of redundancy and noise with simultaneously extraction of meaningful components in multiframe sensor data. Most of these algorithms use a 2D transform of an image to reduce redundant spatial information and different methods of motion compensated difference coding in the time dimension. NETROLOGIC's main innovation is to apply wavelet transform in all three dimensions to reduce the computational complexity while achieving precise and accurate object data extraction in real time. We propose to develop a wavelet based 3D video processing method which combines simultaneous spatial and temporal domain predictions in an attempt to achieve stable occluded target recognition in diverse environment. There are many potential Commercial and Military applications of this technology. The ability to develop compact non-redundant presentation of image data with further model-based classification can result in both a space saving, a reduction in required RAM, increased accuracy in object recognition systems, computer tomography, feature extraction from medical and satellite imagery.
Keywords: Phase_I, NASA, Abstract, FY94
