For　the　purpose　of　mining　distributions　and　movement　patterns　of　moving　objects　at　different　levels　of　details,　this　paper　presents　a　semantic-based　multi-scale　trajectory　modeling　and　knowledge　discovery　method.　Two　matrixes　are　used　to　represent　the　multi-scale　trajectory.　A　matrix　based　frequent　item-set　mining　algorithm　is　developed　to　discover　the　association-rules　in　trajectory　database.　Experiments　show　that　the　approach　is　correct　and　feasible.