Robust tracker of small, fast-moving low-contrast targets

We present a multiresolution adaptive wavelet transform to locate small low-contrast targets. Our approach expands upon methods which use adaptive filters to remove noise to produce a near real-time robust tracker using a specially adapted Kalman filter. This generates a small set of hypotheses to test. Incorrect hypotheses are removed using an interest operator founded on the error covariance generated by the Kalman filter. We demonstrate the technique using some experimental results.