To　avoid　failing　in　visual　tracking　situation　when　employing　single　feature,　a　nonlinear　target　tracking　method　based　on　multi-feature　fusion　is　proposed.　Grey　histogram　is　used　to　describe　the　overall　distribution　characteristics　of　the　target　and　edge　feature　is　employed　to　extract　the　high　frequency　details.　The　two　algorithms　are　fused　in　the　probabilistic　model　of　particle　filter.　Feature　reliability　estimation　based　on　half-band　width　and　contribution　is　proposed,　which　provides　more　reliable　features　with　more　particles.　In　this　way,　the　particle　numbers　of　the　features　are　adjusted　dynamically.　Compared　with　single-feature　tracking　method,　the　tracking　result　shows　that　the　algorithm　has　the　strong　ability　of　tracking　under　local　obstruction.　The　average　tracking　error　of　the　new　algorithm　decreases　by　0.5　pixels.