Big　GPS　data　from　the　moving　vehicles　allow　understanding　the　transportation　systems　in　more　details　and　at　a　vehicle　level.　While　only　depending　on　the　trajectory　and　speed　data　from　probe　vehicles　cannot　estimate　the　traffic　states　(vehicle　densities　etc.)　in　real-time　with　the　classical　filtering　methods,　such　as　particular　filtering　(PF)　and　extended　Kalman　filtering　(EKF),　because　the　boundary　fluxes　of　the　considered　road　network　are　not　available　in　advance.　In　this　paper,　we　firstly　build　the　traffic　flow　dynamics　of　freeway　traffic　as　the　discrete-time　state-space　models　including　unknown　input　variables　(boundary　fluxes).　Then　a　novel　simultaneous　state　and　input　estimation　technique　is　developed　by　using　the　decentralized　heterogeneous　data　of　vehicle　speed　as　measurement.　A　freeway　link　of　Interstate　80　East　(I-80E)　in　Emeryville,　Northern　California,　is　chosen　to　investigate　the　performance　of　the　developed　filtering　approach.　Traffic　data　are　obtained　from　the　Performance　Measurement　System　(PeMS)　and　the　Mobile　Century　experiment.　©　2017　IEEE.