%0 Journal Article
%A JIANG Lin
%A ZHAO Huan
%A XIE Xiao-yan
%T Dynamic Gesture Recognition Based on Characteristics of Encoded Video Data
%D 
%R 10.13190/j.jbupt.2020-056
%J Journal of Beijing University of Posts and Telecommunications
%P 91-97
%V 43
%N 5
%X Aiming at the challenges to scene adaptability and computational complexity of dynamic gesture recognition, a method based on characteristics of encoded video data is proposed. Firstly, density-based spatial clustering of applications with noise is used to extract motion trend features from motion vectors. Then, the motion trends are classified by random forest. Finally, combined by the hand shape features extracted by convolutional neural network(CNN), the dynamic gesture is recognized. The experiment shows that the proposed method has an average recognition rate of 94.22% and 94.48% respective for university of Cambridge and Northwestern University hand gesture data sets. Compared with the scheme combine of CNN and long short-term memory, the gesture recognition time is reduced by 85%. It can still maintain a higher recognition rate for the complex background with insufficient illumination, represents a higher robustness.
%U https://journal.bupt.edu.cn/EN/10.13190/j.jbupt.2020-056