Logo image
以少量的感知器重建數位人動作之研究
Thesis

以少量的感知器重建數位人動作之研究

曾尹璇
Masters, National Tsing Hua University
2010

Abstract

智慧型工廠數位模擬動作擷取動作重建動作預測類神經網路動作資料庫數位人模型 smart factorydigital simulationmotion capturemotion reconstructionmotion predictionartificial neural networkmotion databasedigital human model
In digital factory simulation and computer animation, the issues of human motion are complex and become more and more important. To make the process of human motion generation efficient and cost effective, this paper proposes a method to reconstruct the whole-body motion with fewer sensors in motion capture system.The main part is to construct the human motion reconstruction model by using a motion database. The first step is to select the key data of joints by PCA (principal component analysis). Then the selected data are taken as the input to train the artificial neural network model. The trained model can predict the data of the remaining joints with the fewer sensors on the subjects.The results show that we can use the trained model to reconstruct the human motion well by using partial joint data with fewer sensors. The average reconstruction error of the model for single motion type with information of 6 features is about 1 degree per joint angle. It is similar to the previous works. The main difference is that the method can be easily applied to various motion types and have good performances. The average reconstruction error of the model for various motion types with 10 sensors is also about 1 degree per joint angle.Additionally, this paper provides a simple and fast approach to integrate motion capture system, motion reconstruction model and different motion databases. All the exported motion data can be easily used to animate the digital human model in virtual environment.

Metrics

1 Record Views

Details

Logo image