In　this　paper,　an　attitude　estimation　algorithm　based　on　modified　two-stage　Extended　Kalman　Filter　(EKF)　is　presented　to　improve　the　accuracy　of　attitude　estimation　based　on　Micro　Electro-Mechanical　System　(MEMS)　motion　capture　technology.　Firstly,　a　two-stage　EKF　is　presented　to　replace　the　conventional　EKF,　to　reduce　the　computational　complexity　and　computing　time.　Secondly,　the　normalized　quaternion　model　is　applied　instead　of　Euler　Angle　model　to　reduce　the　calculation　error　of　attitude　estimation.　Finally,　the　automatic　error　compensation　realized　through　constructing　the　acceleration　error　covariance　operator　to　adjust　error　covariance　matrix　according　to　the　measurement　values　of　the　acceleration.　The　experiment　result　shows　that　the　proposed　method　can　significantly　reduce　the　computational　complexity　and　the　size　of　universal　joint　deadlocks　and　linear　acceleration　effect　on　the　attitude　estimation.