Dynamic disparity adjustment and histogram-based filtering of range data for fast 3-D hand tracking

Range data is very important in human-computer interaction applications. Although less costly, range acquisition and processing still presents a speed vs data reliability tradeoff. This paper proposes a method that, given noisy and generally unreliable range data, can filter out erroneous information using range histograms for regions of interest selected in registered color data. Then, using the resulting consistent data that has passed filters, this method limits the depth search space dynamically using motion history and its current state. Experimental results demonstrate the success of the proposed algorithm. Using filtered range data, the algorithm correctly identified the hand involved in manipulation 99.85% of the time. Dynamic disparity adjustment produced a speedup of 60.17% over a static disparity range selection. An application to virtual reality navigation is also presented.