The detection of burst-type events in the output of ground gravitational wave detectors is particularly challenging. The potential variety of astrophysical waveforms, as proposed by simulations and analytic studies in general relativity and the discrimination of actual signals from instrumental noise both are critical issues. Robust methods that achieve reasonable detection performances over a wide range of signals are required. We present here a hybrid burst-detection pipeline related to time–frequency transforms while based on matched filtering to provide robustness against noise characteristics. Studies on simulated noise show that the algorithm has a detection efficiency similar to other methods over very different waveforms and particularly good timing even for low amplitude signals: no bias for most tested waveforms and an average accuracy of 1.1 ms (down to 0.1 ms in the best case). Time–frequency-type parameters, useful for event classification, are also derived for noise spectral densities unfavourable to standard time–frequency algorithms.