Internet　of　Vehicles　(IoV)　based　on　crowdsensing　technology,　which　gets　traffic　data　by　smartphone　or　panel　PC　from　ordinary　person,　has　solved　the　problem　that　getting　sufficient　data　at　low　cost.　However,　it　also　makes　a　new　problem　that　the　data　quality　of　the　system　is　deteriorated.　To　solve　this　problem,　by　analyzing　the　structure　of　crowdsensing　data　and　the　characteristics　of　abnormal　data　in　crowdsensing　IoV,　a　data　detection　algorithm　is　put　forward　to　eliminate　the　abnormal　data　in　IoV　system　and　consequently　improve　data　quality.　In　the　algorithm,　kernel　density　estimation　theory　is　used　to　estimate　the　probability　density　of　traffic　data,　and　a　belief　function　is　then　constructed　to　derive　the　confidence　value　of　every　detected　data.　According　to　the　statistical　theory,　the　data　whose　confidence　value　is　less　than　0　is　regarded　as　abnormal　data.　Finally,　the　feasibility　and　performance　of　the　presented　algorithm　are　simulated.　The　results　show　that　the　proposed　algorithm　can　meet　practical　demands　and　achieve　better　performance　than　that　of　traditional　statistical　detection　methods.　©　2017,　Editorial　Department　of　Journal　of　Hunan　University.　All　right　reserved.