Parking　stall　detection　is　one　of　the　most　important　techniques　of　parking　lot　information　management　system:　the　purpose　of　the　system　is　to　acquire　the　information　about　whether　each　parking　stall　is　occupied　or　not.　It　is　helpful　not　only　for　the　supervision　of　the　manual　management　but　also　for　the　parking　guidance　system.　This　paper　proposes　a　new　algorithm　of　parking　stall　detection.　which　based　on　video.　In　the　first　stage.　this　algorithm　extracts　multiple　static　image　features　of　parking　stall　region.　just　as　variance.　correlation　coefficient.　and　density　of　edge　points.　Then　a　Fisher　decision　is　developed　to　discriminate　whether　there　is　vehicle　on　each　parking　stall.　The　advantages　or　this　algorithm　are　that　the　system　is　robust　to　noise　and　has　high　detection　rate.　The　proposed　algorithm　was　tested　on　a　roadside　parking　lot　for　1687　images　with　different　weather　conditions.　The　result　confirmed　the　efficiency　of　the　proposed　algorithm,　with　the　low　false　alarm　rate　(3.67%)　and　high　detection　rate　(98.7%).