A multiresolution image matching algorithm for moving object detection by wavelets

In the algorithm, images are at first transformed into wavelet representations by means of the Harr wavelet decomposition. Multiresolution feature images are then abstracted from the pyramidal detail images of the wavelet representations. Structural relationships among the objects of interest in a scene are described by means of the binary graphs of the features. Moving object detection is achieved by matching two sequential images of the moving objects. The matching process is started with normalized correlation of grey image patches on the low frequency approximations of the original images in the wavelet structures and kept on matching the graphs on the feature pyramids. Experiments show that the method is more efficient and robust than traditional methods