Target extracting algorithm for complex scene in vehicle image tracking system

A new target extracting algorithm for complex scene is proposed. An improved Snake model is presented and applied to the construction of initial template, and the fractal Brown random field model is introduced to work out the possible target region according to the wavelet fractal dimension and fractal fitting error. After that, a new minimum mismatch distance (MMD) similarity measurement is defined, and the correlation matching is made quickly based on the feature region of the target. By accurately constructing initial template and using coarse-to-precise target searching strategy, the algorithm guarantees good extraction precision and reduces the computation greatly.