Technology of human motion capture has been widely used in digital entertainment field. Editing the existing large amount of human motion capture data, correcting and eliminating motion distortion caused by noise and ...
详细信息
Technology of human motion capture has been widely used in digital entertainment field. Editing the existing large amount of human motion capture data, correcting and eliminating motion distortion caused by noise and other defects have important value and significance for data reuse. In this paper, data processing is carried out based on convolutional automaticencoder and manifold learning. The popular structure of human motion data was learned by a one-dimensional time domain convolution automatic encoder, in which the hidden unit of the automaticencoder represents motion data. Three constraints were used to overcome the problem that the hidden unit has too much motion editing range. The data to be processed in this paper has no limit on the number of motions. The proposed method can process large data sets in parallel and automatically perform manifold learning without manual labelling and segmentation. In the final, comparative experiments based on a variety of damaged motion data have been carried out. The results showed that the proposed method can effectively reduce the error of the original motion data, and has achieved good results in both objective evaluation and subjective evaluation.
暂无评论