Abstract: The objective of motion capture data denoising aims to recover the frame sequences to better express the origi-nal data characteristics from the noise corrupted motion capture data．In general,the frame sequences within a short period of human motion capture data always reflect the same or similar motion semantic behavior．To this effect,this paper presented a Segmented Low Rank Approximation (SLRA)approach for motion capture data denoising．The proposed approach first divides the noise corrupted motion sequence into several continuous subintervals．Then,the inexact augmented lagrange multiplier method (IALM)is employed to decompose each subinterval batch matrix in terms of the low rank matrix approximation and sparse noise error estimation．Accordingly,the noise corrupted information within each frame subinterval can be removed．Finally,all the approximated low rank matrixes corresponding to the segmented subintervals are sequentially combined to represent the whole recovered sequence from the noise corrupted motion data．The simulation experimental results show that the proposed approach is able to well perform denoising of the human motion capture data with arbitrary topologies．The satisfactory performance demonstrates its universality and practicality.

Key words: Motion capture data denoising,Segmented low rank matrix approximation,Continuous subinterval,Inexact augmented lagrange multiplier,Sparse noise error