Over　the　past　years,　automatic　traffic　accident　detection　(ATAD)　based　on　video　has　become　one　of　the　most　promising　applications　in　intelligent　transportation　and　is　playing　a　more　and　more　important　role　in　ensuring　travel　safety.　This　paper　proposes　a　classifier-based　supervised　method　by　viewing　the　last　seconds　before　motor　vehicle　collisions　as　the　detection　target.　In　our　method,　we　devise　a　novel　algorithm　called　OF-SIFT　as　the　low-level　feature.　Deriving　from　the　optical　flow　and　Scale　Invariant　Feature　Transform　(SIFT),　it　is　designed　to　extract　local　motion　information　from　the　temporal　domain　rather　than　gradient-based　local　appearance　from　the　spatial　domain.　The　purpose　of　OF-SIFT　is　to　generate　a　feature　that　can　capture　sufficient　and　distinctive　dynamic　motion　information　for　motion　detection　without　using　the　static　state　information　of　moving　objects.　Further,　in　order　to　develop　a　more　compact　image　representation　without　considering　the　explicit　vehicle　geometry　shape,　we　use　the　idea　of　Bag　of　Feature　(BOF)　model　with　spatial　information　to　encode　features.　Finally,　an　extreme　learning　machine　(ELM)　classifier　is　introduced　as　the　basic　classifier　owing　to　its　excellent　and　fast　generalization.　Experiments　using　real-world　data　have　shown　that　the　proposed　method　has　achieved　good　performance　in　handling　ordinary　video　scenes.　©　2016　IEEE.