Tracking Small Hand Movements in Interview Situations

In this paper, we motivate ongoing work into developing methods for the automated tracking of small hand movements in interview situations to aid nonverbal behaviour analysis in the detection of deception. Existing techniques for detecting and tracking hand motion are reviewed to place current and future technical work into context. We present a modification to the popular colour predicate approach to skin detection based on Bayesian posterior probability maps and Parzen colour space probability density estimates. We demonstrate the application of a complex wavelet decomposition to identify changes in finger position. Although our existing hand tracking algorithm currently relies on posterior probability map thresholding, morphological operations and distance heuristics, we suggest the role of kinematic models of upper body, limb and finger motion for future work.