As　the　basic　function　of　advanced　assisted　driving　technology　or　autonomous　driving　technology,　vehicle　recognition　plays　an　important　role　in　lane　change,　collision　avoidance　and　other　applications.　The　current　vehicle　recognition　method　has　the　problem　of　slow　recognition　speed　and　false　detection　caused　by　camera　movement.　In　order　to　solve　this　problem,　this　paper　improves　the　existing　three-frame　difference　method.　The　candidate　frames　obtained　by　the　three-frame　difference　method　are　then　screened　and　identified　by　FAST　feature　extraction,　and　tested　in　a　typical　vehicle　video　data　set.　The　test　results　show　that　the　average　number　of　targets　detected　by　mistake　per　frame　is　reduced　from　1.5　to　0.15　when　the　camera　is　rotating　incorrectly,　and　the　detection　rate　is　also　increased　by　10%　when　the　camera　is　not　moving.　©　2024　IEEE.