Aimed　at　solving　the　problems　of　GPS　outage　or　deterioration　of　accuracy,　the　strap-down　inertial　navigation　system　(SINS)　is　proposed　to　provide　reliable　positioning　or　attitude　information　for　intelligent　vehicle.　A　set　of　mathematical　transformations　and　integrations　with　respect　to　time　are　applied　to　these　measurements　of　SINS　to　calculate　position,　velocity　and　attitude　information.　More　importantly,　in　allusion　to　the　deterioration　with　time　of　SINS　accuracy,　caused　by　the　possible　noise,　this　paper　analyzes　the　stochastic　mathematical　model　of　SINS,　uses　two　de-noising　methods　to　preprocess　the　raw　experimental　data　of　the　intelligent　vehicle　for　eliminating　integrated　error.　By　comparing　the　position　errors　calculated　with　three　sets　of　data:　the　raw　data,　the　wavelet　de-noised　data　and　the　adaptive　fading　Kalman　filtered　data　based　on　the　current　statistical　model　of　acceleration　in　the　experiment,　we　draw　a　conclusion　that　inertial　navigation　algorithm　with　data　preprocessed　by　the　two　de-noising　methods　can　improve　the　position　accuracy　considerably.　And　the　proposed　adaptive　fading　Kalman　filter　method　based　on　the　current　statistical　model　of　acceleration　is　better　than　the　wavelet　de-noising　method.