To　realize　a　human　tracking　task　in　a　cluttered　environment,　a　method　of　multi-feature　based　human　tracking　under　the　CTF(coarse-to-fine)　locating　strategy　was　proposed.　The　proposed　method　located　the　target　from　a　RFID　system　coarsely.　Then,　the　silhouette　of　the　head-shoulder,　the　cloth　color　and　motion　feature　were　extracted　to　locate　the　target　accurately　by　using　the　processing　techniques　including　adaptive　template　matching　algorithm,　improved　Camshift　and　Extended　Kalman　Filter.　At　last,　an　intelligent　gear　shift　controller　based　on　fuzzy　rules　considering　the　motion　state　of　the　target　and　the　robot　was　utilized　to　drive　the　robot.　The　experimental　results　show　that　the　presented　method　can　keep　the　robot　in　a　suitable　distance　from　the　target　and　handle　the　problem　of　occlusion,　a　sudden　turn,　and　complicated　background.