Traditional　fire　detection　methods　are　based　on　smoke　and　detectors.　They　are　not　suitable　for　high　and　large-span　space　structures　because　of　their　limited　detection　range.　The　latest　fire　detection　methods　are　based　on　video-image　processing　and　data　fusion.　However,　false　positive　rate　and　false　negative　rate　still　remain　unsatisfactory　and　need　improvement.　In　this　paper,　some　fire　video-image　detection　algorithms　are　studied.　A　prototype　system　is　developed　to　verify　the　performance　of　these　algorithms.　A　series　of　algorithm　tests　on　fire　video　file　are　conducted.　It　is　found　that　detection　algorithms　on　the　basis　of　fuzzy　neural　network　behave　more　fine　than　those　based　on　probability　density,　historical　data　fusion　can　lower　false　positive　rate　and　false　negative　rate　remarkably,　it　is　not　true　that　evidence　combination　rules　(Dempster-Shafer　rules)　can　always　get　a　more　satisfying　fusion　result.