Many astronomical sources produce transient phenomena at radio frequencies,but the transient sky at low frequencies (<300 MHz) remains relativelyunexplored. Blind surveys with new widefield radio instruments are settingincreasingly stringent limits on the transient surface density on varioustimescales. Although many of these instruments are limited by classicalconfusion noise from an ensemble of faint, unresolved sources, one can inprinciple detect transients below the classical confusion limit to the extentthat the classical confusion noise is independent of time. We develop atechnique for detecting radio transients that is based on temporal matchedfilters applied directly to time series of images rather than relying onsource-finding algorithms applied to individual images. This technique haswell-defined statistical properties and is applicable to variable and transientsearches for both confusion-limited and non-confusion-limited instruments.Using the Murchison Widefield Array as an example, we demonstrate that thetechnique works well on real data despite the presence of classical confusionnoise, sidelobe confusion noise, and other systematic errors. We searched fortransients lasting between 2 minutes and 3 months. We found no transients andset improved upper limits on the transient surface density at 182 MHz for fluxdensities between ~20--200 mJy, providing the best limits to date for hour- andmonth-long transients.