Research On Editing And Reuse Technology Of Human Motion Capture Data

Human motion capture technology records the human body's trajectory in three-dimensional space accurately.And it is widely used in films,virtual reality,sports and other fields.The popularity of human motion capture technology makes how to edit and reuse a large amount of motion capture data effectively become a challenging topic.The editing and reuse processing of motion capture data contains a series of related technologies.This paper studied the three important technologies of them,which are human motion segmentation,key-frame extraction and human motion synthesis.The main contents of this paper include:In the aspect of human motion segmentation,this paper proposed a motion segmentation model based on multiple information of motion capture data.The model utilizes both low-level physical information and high-level data distribution information of motion capture data for motion segmentation.Firstly,introducing the Density Peaks Clustering algorithm to solve the problem of motion segmentation,and optimizing it to obtain a better initial segmentation result.Then,using the Aligned Cluster Analysis method to optimize the segmentation result of initial segmentation.The optimization algorithm of Density Peaks Clustering and Aligned Cluster Analysis method complement each other,which improves the performance of the motion segmentation model in terms of accuracy and stability.In the aspect of key-frame extraction,this paper proposed a key-frame extraction method based on the Adaptive Quantum-Behaved Particle Swarm Optimization algorithm.This method uses the Swarm Intelligent Optimization algorithm to solve the key-frame extraction problem.Firstly,introducing the Quantum-Behaved Particle Swarm Optimization algorithm and improving its adaptive ability using the Cloud Model.Then,searching for key-frames in the motion capture data to obtain the key-frames with the smallest reconstruction error.The improved global convergence and optimization ability of the Adaptive Quantum-Behaved Particle Swarm Optimization algorithm ensures that the extracted key-frames have good performance in motion abstract.In the aspect of human motion synthesis,this paper proposed a motion style transfer model based on convolutional autoencoder.From the perspective of motion style,the model applies a deep learning network to human motion synthesis,and transfers the human motion style through convolutional autoencoder.Firstly,constructing and training a convolutional autoencoder to extract motion features.Then,creating the style transfer loss by the convolutional autoencoder and calculating the motion data of the hidden layer according to the style transfer loss.Finally,the motion sequence after motion style transfer by the convolutional autoencoder is generated.