Human Motion Capture Data Recovery Using Temporal Difference Low-rank Constraint

Recovering missing data from observed human motion capture data is an important research issue.During data collection,poor lighting at the collection site,marker points being blocked by other objects or by the collector’s own body parts often occur,resulting in motion distortion.Therefore,recovering damaged motion data into accurate and natural motion data is very necessary and challenging.The main idea of current human motion capture data recovery methods based on low-rank matrix completion is to use the lowrankness,noise sparsity,and temporal stability of human motion data to remove noise and estimate missing markers.This paper proposes two motion capture data models based on the low rank temporal difference of human motion capture data.The specific research results are as follows:(1)A human motion capture data recovery algorithm using convex temporal difference low-rank constraint is proposed.In order to better characterize the temporal low-rank property of data,first,the data matrix is projected into a temporal difference space,and then a convex kernel norm is introduced to impose low-rank constraint on the differential data.On the one hand,the model considers both the low rank property and temporal stability of the data,simplifying the vodel,and on the other hand,effectively reducing the impact of sparse noise on motion capture data.To solve the model,the alternating direction method of multipliers and(inverse)discrete cosine transform are used to quickly solve the model.Finally,a comparative experiment is conducted between this algorithm and classical algorithms,by comparing the restoration error and visual restoration effect,the results show that the algorithm proposed in this paper can effectively restore corrupted data.(2)A non-convex temporal difference low-rank constraint human motion capture data recovery algorithm combining Schatten-p norm and lq norm is proposed.Firstly,Firstly,a temporal difference matrix is constructed,and a non-convex Schatten-p norm is introduced to characterize the temporal low-rank property of a data matrix and introduce a non-convex lq norm to constrain sparse noise terms.Then,the alternating direction method of multipliers is used to solve the corresponding subproblem.When solving the corresponding subproblem,the Newton iterative method is used to solve it and a convergent solution can be obtained.Finally,a large number of experimental data comparison proves that this method has better recovery performance.