In　order　to　extract　numbers　from　an　aerial　video,　an　improved　algorithm　of　the　text　event　detection　is　presented　in　this　paper.　This　method　intercepts　key　frames　using　the　way　of　comparison　between　frame　difference　and　the　dynamic　threshold.　In　this　way,　it　can　avoid　unnecessary　trouble　made　by　dealing　with　the　video　frame　by　frame.　Results　show　that　this　method　can　save　much　computing　time　as　well　as　improve　operating　efficiency.　Besides,　a　new　method　of　de-noising　for　binary　image　is　proposed　during　image　preprocessing.　For　the　sum　of　the　column　pixel　of　noise　is　much　less　than　the　concerned　area,　we　can　successfully　remove　the　noise　pixel　and　pave　the　way　for　the　future　steps　such　as　character　division　or　number　recognition.　As　the　results　shown,　this　method　is　especially　suitable　for　processing　poor　quality　images.　©2009　IEEE.