ITR: Providing Intuitive Access to Human Motion Databases

我们捕捉人类运动数据的能力已经大大提高，使人类运动数据库能够支持创造性和教育性的努力，如运动员的训练，诊断和治疗运动障碍，并保护我们的文化遗产。然而，为了产生影响，运动数据库必须能够被广泛的受众访问，包括没有编程经验的儿童和成人用户。在这个项目的时间框架内，我们将对人类运动数据库的基于性能的界面进行探索。该界面将允许用户执行所需的动作，并将需要从性能中比较时间序列数据，可能有丢失或嘈杂的条目，到数据库中的时间序列数据。检索技术必须计算样本和数据库中任意运动子序列之间的距离，这使得蛮力技术非常昂贵，特别是在交互性能方面。将实现降维、特征转换和自动分割的算法，将定量地和通过用户研究比较距离函数，并评估这些算法在缺少数据时的性能。该研究将通过传统的数据库检索评估技术以及对界面用户的观察和分析进行评估。

Our ability to capture human motion data has increased dramaticallymaking possible human motion databases that can support creative andeducational endeavors such as training of athletes, diagnosing andtreating motion disorders, and preserving our cultural heritage. Tohave an impact, however, motion databases must be accessible to a wideaudience, including children and adult users with little programmingexperience. During the time frame of this project, we will undertakean exploration of a performance-based interface for a human motiondatabase. The interface will allow the user to act out the desiredmotion and will require comparing time sequence data from the performance,perhaps with missing or noisy entries, to time sequence data in thedatabase.Retrieval techniques must compute the distance between a sample andarbitrary motion subsequences in the database, making brute forcetechniques prohibitively expensive, particularly for interactiveperformance. Algorithms for dimensionality reduction, featuretransformations, and automatic segmentation will be implemented,distance functions will be compared both quantitatively and via userstudies, and the performance of these algorithms with missing data willbe assessed. The research will be evaluated with traditional databaseretrieval evaluation techniques and through observation and analysis ofusers of the interfaces.