摘要： 针对根据关键帧重建运动序列时重建误差率较高的问题,提出一种基于改进的t-SNE降维算法,提取人体运动数据关键帧。利用t-SNE对原始运动数据进行降维,通过改变t-SNE中函数宽度参数的计算方法,得到稳定的低维特征曲线。提取低维特征曲线的局部最大值、最小值作为初始关键帧,再使用曲线幅度算法提取最终的关键帧序列。实验结果表明,在相同压缩比下,与其他关键帧提取方法相比,该方法具有较低的重建误差率。

关键词: 运动捕捉, 改进的t-SNE算法, 特征曲线, 关键帧提取, 曲线幅度

Abstract: Aiming at the problem of high error rate in the reconstruction of motion sequences according to key frames,this paper proposes an improved t-SNE dimension reduction algorithm to extract key frames of human motion data.The proposed t-SNE algorithm is used to reduce the dimension of original motion data.A low dimensional characteristic curve can be obtained by changing the computing method of function width parameter for t-SNE.The local maximum and minimum values of the low dimensional characteristic curve are adopted as the initial key frames,and the final key frame sequence can be extracted based on the curve amplitude algorithm.Experimental results show that under the same ratio of compression,compared with other key frame extraction methods,the proposed method has lower reconstruction error ratio.

Key words: motion capture, improved t-SNE algorithm, characteristic curve, key frame extraction, curve amplitude

中图分类号: 

• TP37