Spatio-temporal interest region detectors can be used in the analysis of video to determine sparse, informative regions as candidates for feature extraction. In this paper we compare existing detectors and introduce the new FAST-3D detector, loosely based on the FAST spatial interest region detector. We compare the invariance of detectors to rotation, scale and compression by measuring the similarity between detected interest regions in original and transformed versions of videos. We measure both the repeatibility and introduce a new similarity measure based on mutual information. The FAST-3D detector is shown to be on par with the other detectors, while showing a significant increase in speed.