预选策略和重建误差优化的运动捕获数据关键帧提取

Extraction of Key-frame from Motion Capture Data Based on Pre-selection and Reconstruction Error Optimization

为了从运动捕获数据中提取关键帧来进行人体动画创作,提出一种以重建误差或压缩率为目标的运动捕获数据关键帧提取方法.该方法将关键帧提取划分为帧预选和基于重建误差优化的精选2个阶段,在第一阶段,对原始动作序列筛选边界姿势作为候选关键帧,使最终关键帧序列具有较强运动类型描述能力;在第二阶段,定义帧消减误差作为关键帧重要性的度量标准,并定义最大重建误差作为关键帧提取过程中的优化目标,同时考虑压缩率目标,精选候选帧获得满足指定重建误差或压缩率要求的最终关键帧集合.实验结果表明,文中方法具有良好的数据压缩效果,能满足实时压缩的需要.

In order to extract keyframes for animation creation, a method aiming at increasing compression rate and decreasing reconstructed error is proposed. The proposed method consists of two phases, that is, the pre-selection phase and the refinement phase. In the first phase, extremal postures in original motion capture data curve were pre-selected as candidate keyframes. In the second phase, firstly the decimated error was introduced as a metric to measure the importance of keyframe and the reconstructed error was introduced as an optimal objective in the keyframe extraction. In order to get the result keyframes, two algorithms aiming to specified compression rate and reconstructed error were proposed to eliminate less important candidate frames to refine the candidate keyframes, where both compression rate and reconstructed error were optimized. The experimental results demonstrate that the original motion capture data can be compressed at a high ratio, and the proposed approaches meet the requirement of real-time data compression.