Over the past years, several schemes have been proposed to estimate the perceptual distortion in video quality caused by packet loss. Some of them are intuitively designed and do not correlate well with actual distortion values in most real-world scenarios. Others simulate decoders' error concealment measures and use the quality degradation of reconstructed regions as basis for calculating importance estimates. However, such techniques do not take error propagation in the temporal domain into account and are computationally expensive, especially when estimates cannot be derived in parallel to, or as side-product of the video encoding process. We therefore propose a novel approach which targets all previously mentioned problems and inspects dependencies between media units at the level of macroblocks. We show how our scheme can be applied to live video streaming systems using rateless codes for error protection and identify pitfalls which have to be considered. Test results indicate that the proposed unequal error protection scheme considerably outperforms previous approaches, independent of the resolution of the test sequences.