Multi-resolution Wavelet Image Tracking

The introduction of a whole new class of transform methods, notably the wavelet transforms, which permit decomposition of observed phenomena into different physical scales and the computational leverage provided by multiscale image processing tech- niques, when applied to multisensor Lmage data, promises to increase the sensitivity and reliability of Lmage processing systems in several applications. This proposal offers to evaluate the feasibility of multiscale statistical methods, generalizing wavelet transform methods, for solving two key problems of many Lmage tracking systems: identification of textured background and estimation of optical flow based on sequences of images. The products would be software Lmplementation of algorithms for the two problems and an analysis of their perfor- mance using sLmulated and video images. The images would represent one optical sensor viewing a single target moving in a structured, tLme-varying background.