Indexing Large Human-Motion Databases

Abstract

Data-driven animation has become the industry standard for computer games and many animated movies and special effects. In particular, motion capture data recorded from live actors, is the most promising approach offered thus far for animating realistic human characters. However, the manipulation of such data for general use and re-use is not yet a solved problem. Many of the existing techniques dealing with editing motion rely on indexing for annotation, segmentation, and re-ordering of the data. Euclidean distance is inappropriate for solving these indexing problems because of the inherent variability found in human motion. The limitations of Euclidean distance stems from the fact that it is very sensitive to distortions in the time axis. A partial solution to this problem, Dynamic Time Warping (DTW), aligns the time axis before calculating the Euclidean distance. However, DTW can only address the problem of local scaling. As we demonstrate in this paper, global or uniform scaling is just as important in the indexing of human motion. We propose a novel technique to speed up similarity search under uniform scaling, based on bounding envelopes. Our technique is intuitive and simple to implement. We describe algorithms that make use of this technique, we perform an experimental analysis with real datasets, and we evaluate it in the context of a motion capture processing system. The results demonstrate the utility of our approach, and show that we can achieve orders of magnitude of speedup over the brute force approach, the only alternative solution currently available. KeywordsMotion Capture, Animation, Time Series, Indexing INTRODUCTIONData-driven animation has now become the industry standard for the production of computer games and many animated movies and special effects. The most promising and widely applied approach so far is the use of motion capture data. These are motion data recorded from live actors, which can subsequently be used for animating realistic human characters. Motion capture data, in its rawest form, is recorded with a few technologies, the most popular of which appears to be optical (see Vicon [38] and Motion Analysis [39] products) in which digital cameras record small reflective markers fixed to the human actor as he/she moves. Through multiple cameras and triangulation, three dimensional position traces for the markers are resolved faithfully. The markers can then be identified (as outer left knee, for example) and filtered. Motion capture allows the animation of a 3D model, where the data is mapped to the skeleton of the desired character and body orientations are determined ( Figure 1). In practical applications, most motion capture data is stored in segmented sequences in a motion library, for example a modern sports game may contain thousands of motion data "clips". The system, i.e. game engine in this case, selects and plays motions from the database [37]. Our approach aids in the creation and manipulation of such libraries by quickly finding instances of a given motion se...

Boundary-based lower-bound functions for dynamic time warping and their indexing

“…For example, subsequence matching under DTW is considered in [36]; uniform time scaling matching under DTW is discussed in [10,4]; the non-uniform time scaling matching under DTW is attacked in [39], where the authors propose using a segment-based rather than point-based DTW algorithm to adapt to the time-scaling scenario.…”

Adaptive multi-view feature selection for human motion retrieval

“…In general, the main motion retrieval procedure includes three steps: (1) visual feature extraction; (2) training a search/ranking model; (3) similar human motion retrieval or ranking. To speed up the searching procedure, some data indexing techniques such as kd-tree and R-tree are used in approximate searching [17][18][19]. Indeed, the success of human motion retrieval heavily relies on the motion feature representation.…”

One Way Distance: For Shape Based Similarity Search of Moving Object Trajectories

“…But this algorithm can only compare trajectories with the same lengths or with the same time interval. [10] proposes a method which allows global stretch to match sequences similar but in different scale. [6] introduces EDR (edit distance on real sequence) distance function based on edit distance.…”

“…Recently interests of similarity search also arise in trajectory comparison [1], [2], [6], [10]- [13], [16], [18]- [20]. DTW has some variants for trajectories similarity search [12], [18].…”

“…Similarity search is not a new topic and has been investigated in various context, e.g., motion tracking in videos [10], [12], time series analysis [4], [5], [8], [9], [14], [16], [17], and recently, trajectories, [6], [11], [13], [16], [18]- [20]. Partly due to the difficulty in formalizing "similarity" and partly due to the diversity of applications, known results are not satisfactory in many application contexts.…”