A Personalized Gesture Interaction System with User Identification Using Kinect

Are mid-air dynamic gestures applicable to user identification?

“…Recently, mid-air dynamic gesture based biometrics is proposed in works [8,9,10]. The basic principle of this biometrics means is that even different individuals draw the same type gestures, there will be inherent motion differences due to behavioral habits.…”

“…Most current gesture based user biometrics works [9,10] mainly focus on specific tasks and are under small samples data sets and they still employ the handcrafted features and adopt the traditional methods -DTW (dynamic time warping) or SVM (support vector machine) for user identity matching. Furthermore, these kind of methods distinguish individuals only by comparing the dynamic gesture trajectories without considering the correlation and the relationship between the gesture motion features and the gesture characteristics for user identity.…”

Design and evaluation of novel gesture-centric interaction models in smart spaces

“…Gestures as behavioral attributes for identification are less invasive than the biometric physical ones. Previous works identify users with in-air single-gesture of digits [111], implicit pick-up or shaking [32], and signatures [69]. In this work, user identification based on the Kinect with gesture pairs is proposed, showing in evaluation a significant improvement on accuracy.…”

“…The proposed system can be integrated into gesture-based home entertainment systems with a group size of 3-5 and used for interface customization, content adaption and parental control. To access the tiled-display system for visualization, the users are required to gesture a series of digits as identity passwords in [111]. Wu et al [106] make use of the ability of skeleton tracking of the Kinect to identify users by full body gestures like swing or punch.…”

“…After two lab studies, the system achieves impressive 0.5 to 1.6% ERR in the field study. The identification system CONCLUSION for tiled-display system [111] captures the trajectory movement of single hand joints in the x-y dimension and represents the trajectory of digits as the sequence of movement direction vectors that are quantized. A two-phase cascading classifier combining the classifiers HMM (Hidden Markov Models) and GSS (Golden Section Search) is utilized to mitigate the misclassification problem of pure HMM based solution.…”