TY - GEN
T1 - Hand gesture recognition for a virtual mouse application using geometric feature of finger's trajectories
AU - Maleki, Behnam
AU - Ebrahimnezhad, Hossein
AU - Xu, Min
AU - He, Xiangjian
N1 - Publisher Copyright:
© 2015 ACM.
PY - 2015/8/19
Y1 - 2015/8/19
N2 - We aim to enable a computer to comprehend and perform the mouse functions by analyzing a video with hand motions. For this purpose, dynamic gestures are captured by a web cam and are recognized as pre-defined gestures which are used to suggest mouse functions. The proposed algorithm initially detects the hand. Then, it tracks fingertips' trajectories within a frame sequence. Finally, hand gestures are recognized through computing a set of proposed geometric features of fingers' trajectories and comparing with our collected gestures dataset. In this paper, four types of descriptors are defined for a dynamic gesture. Each descriptor includes different number of features, which compose a feature vector with 135 dimensions. Different classification algorithms (e.g. KNN, LDA, Naïve Bayes and SVM) are applied to compare the detection results. The minimal misclassification error rate (MCR) reaches about 4% (i.e. Correct Recognition rate of 96%). Furthermore, we applied Principle Component Analysis (PCA) to reduce the number of features. With 30 dimensional features (principle components), LDA classifier can achieve about 0.09% misclassification error rate.
AB - We aim to enable a computer to comprehend and perform the mouse functions by analyzing a video with hand motions. For this purpose, dynamic gestures are captured by a web cam and are recognized as pre-defined gestures which are used to suggest mouse functions. The proposed algorithm initially detects the hand. Then, it tracks fingertips' trajectories within a frame sequence. Finally, hand gestures are recognized through computing a set of proposed geometric features of fingers' trajectories and comparing with our collected gestures dataset. In this paper, four types of descriptors are defined for a dynamic gesture. Each descriptor includes different number of features, which compose a feature vector with 135 dimensions. Different classification algorithms (e.g. KNN, LDA, Naïve Bayes and SVM) are applied to compare the detection results. The minimal misclassification error rate (MCR) reaches about 4% (i.e. Correct Recognition rate of 96%). Furthermore, we applied Principle Component Analysis (PCA) to reduce the number of features. With 30 dimensional features (principle components), LDA classifier can achieve about 0.09% misclassification error rate.
KW - Dynamic hand gesture
KW - Gesture classification
KW - Hand tracking
KW - Principle Component Analysis
KW - Recognition
UR - http://www.scopus.com/inward/record.url?scp=84947583187&partnerID=8YFLogxK
U2 - 10.1145/2808492.2808566
DO - 10.1145/2808492.2808566
M3 - Conference contribution
AN - SCOPUS:84947583187
T3 - ACM International Conference Proceeding Series
SP - 16
EP - 19
BT - ICIMCS 2015 - Proceedings of the 7th International Conference on Internet Multimedia Computing and Service
A2 - Jain, Ramesh
A2 - Jiang, Shuqiang
A2 - Smith, John
A2 - Sang, Jitao
A2 - Li, Guohui
A2 - Zhang, Tianzhu
A2 - Wang, Shuhui
PB - Association for Computing Machinery
T2 - 7th International Conference on Internet Multimedia Computing and Service, ICIMCS 2015
Y2 - 19 August 2015 through 21 August 2015
ER -