The impact of motion dimensionality and bit cardinality on the design of 3D gesture recognizers

摘要: The interactive demands of the upcoming ubiquitous computing era have set off researchers and practitioners toward prototyping new gesture-sensing devices gadgets. At same time, practical needs developing for such miniaturized prototypes with sometimes very low processing power memory resources make in high demand fast gesture recognizers employing little memory. However, available work on motion classifiers has mainly focused delivering recognition performance less discussion execution speed or required This investigates today's commonly used 3D under effect different dimensionality bit cardinality representations. Specifically, we show that few sampling points depths are sufficient most metrics to attain their peak context popular Nearest-Neighbor classification approach. As a consequence, 16x faster working 32x while levels being reported. We present results large corpus consisting nearly 20,000 samples. In addition, toolkit is provided assist optimizing order increase reduce consumption designs. deeper level, our findings suggest precision human motor control system articulating gestures needlessly surpassed by sensing technology unfortunately bares direct connection sensors' cost. hope this will encourage consider improving careful analysis representation rather than throwing more into design.