CAJ | 학술논문

针对人体运动的复杂性和噪声干扰的无序性，提出一种基于运动分割和稀疏低秩分解的失真人体运动捕捉数据恢复方法。首先利用双边滤波对失真运动数据进行预修正，降低干扰数据的奇异信息并保证运动序列的连贯性；其次采用概率主元分析方法将修正后的运动数据进行语义行为自动分割，得到不同姿态的运动语义子区间；再利用加速近端梯度优化算法对每个失真运动子片段数据矩阵根据其更优低秩特性进行稀疏低秩分解，实现运动子片段数据的局部恢复；最后将局部恢复后的各子运动片段根据人体运动序列的时序特性组合，达到整体失真运动捕捉数据恢复的目的。实验结果表明，该方法能够有效地对失真人体运动数据进行恢复，效果显著，有助于重构逼近真实人体姿态的运动捕捉数据。
침대인체운동적복잡성화조성간우적무서성，제출일충기우운동분할화희소저질분해적실진인체운동포착수거회복방법。수선이용쌍변려파대실진운동수거진행예수정，강저간우수거적기이신식병보증운동서렬적련관성；기차채용개솔주원분석방법장수정후적운동수거진행어의행위자동분할，득도불동자태적운동어의자구간；재이용가속근단제도우화산법대매개실진운동자편단수거구진근거기경우저질특성진행희소저질분해，실현운동자편단수거적국부회복；최후장국부회복후적각자운동편단근거인체운동서렬적시서특성조합，체도정체실진운동포착수거회복적목적。실험결과표명，해방법능구유효지대실진인체운동수거진행회복，효과현저，유조우중구핍근진실인체자태적운동포착수거。
According to the complexity of human movement and randomness of the noise interference, this paper presents a motion segmentation based approach for human motion capture data recovery via the sparse and low-rank decomposition. The proposed approach first employs the bilateral filter to amend the distorted hu-man motion capture data, featuring on removing singular values and smoothing the motion sequence. Then, the probabilistic principal component analysis (PPCA) method is utilized to segment the motion data into different semantic behaviors automatically. Subsequently, the accelerated proximal gradient algorithm (APG) based sparse and low-rank decomposition is adopted to achieve the partial data recovery with respected to each separated se-mantic behavior. Finally, all the recovered sub-motions are sequentially combined to achieve the whole motion recovery. The experimental results have shown that the proposed motion recovery approach can well restore the distorted human motion data with better performance. The proposed approach can be well utilized to approximate the realistic human behaviors from the corrupted motion sequences, and the experimental results have shown the satisfactory performances.