The　Adaboost　vehicle　detection　algorithm　based　on　Haar　feature　has　a　good　performance　in　real-time　and　accuracy.　However,　the　method　has　a　lot　of　omission　and　error　detection　in　the　detection　of　Special　vehicle　under　complicated　traffic　flow.　In　this　paper,　a　method　of　taxi　window　area　detection　is　proposed　to　replace　vehicle　detection.　At　the　same　time,　a　method　of　sliding　color　histogram　matching　is　proposed　to　reduce　the　error　detection.　Finally,　the　traffic　surveillance　video　is　used　to　verify　the　algorithm,　Detection　results　proves　that　the　algorithm　has　a　good　accuracy　and　real-time　performance　for　the　detection　of　taxi　vehicles.