The MotionExplorer workflow supports domain experts in exploring and searching large collections of human motion capture data (multivariate time series of body poses) to identify motion types, transitions, and style variations for motion synthesis research. The workflow combines hierarchical aggregation with interactive visual exploration and query-by-example search to support overview-first, detail-on-demand analysis of human pose sequences.
The process begins with feature extraction, where high-dimensional motion capture frames are normalised and reduced to compact 48-dimensional pose feature vectors. A divisive hierarchical clustering algorithm then groups similar poses into a binary tree of pose clusters at multiple levels of detail. Two coordinated overview visualisations are generated: a pose hierarchy explorer (dendrogram with cluster glyphs showing centroid stick-figures and variation bands) and a motion explorer (node-link diagram with PCA-projected node positions and temporal transition edges, colour-coded by SOM-based similarity).
The expert enters an iterative exploration loop, inspecting the coordinated views to identify macro-level motion types, pose groupings, and transition patterns. If the current aggregation granularity is inadequate, the expert adjusts the aggregation level via an interactive slider (and optionally filters clusters), triggering re-rendering of both visualisations — this cycle repeats until a satisfactory overview is achieved.
Once an adequate overview is established, the expert proceeds to query formulation, visually selecting start and end pose clusters (via drag-and-drop from cluster glyphs) to define a query-by-example specification. A search and style variation analysis loop follows: matching motion subsequences are retrieved, visualised as colour-coded sequence strips with animation playback, and inspected for style variations (differences in speed, dynamics, path, or intermediate poses). If results are unsatisfactory, the expert refines the query by selecting different start/end poses and the cycle repeats.
Finally, in a knowledge synthesis step, the expert compiles insights about which motion sequences are available for reuse, which style variations exist, and whether additional camera takes need to be recorded.