The　set-membership　Kalman　filtering　is　proposed　to　solve　the　problems　of　the　system　requiring　100%　confidence　to　be　estimated　and　the　real-time　calculation　for　systems　with　nonlinear　equality　constraints.　The　algorithm　of　the　minimum　trace　ellipsoid　is　adopted　to　optimize　the　stage　of　time　updating.　At　the　steps　of　prediction　and　filter,　the　unconstrained　set-membership　Kalman　filtering　is　projected　onto　the　state　constraint　surface　to　deal　with　the　constraint　problem.　The　proposed　algorithm　is　tested　on　a　vehicle　tracking　application.　The　results　show　that　the　true　values　by　using　the　constrained　set-membership　filter　always　reside　between　their　upper　bounds　and　lower　bounds　during　the　later　tracing,　despite　the　original　values　have　great　error　which　rapidly　decreases　compared　to　the　Kalman　filtering.　The　simulation　results　show　that　the　proposed　method　is　available　and　effective.