Deep Learning based Dynamic Hand Gesture Recognition with Leap Motion Controller
International Journal of Advanced Trends in Computer Science and Engineering · 2020
Dynamic hand gesture recognition (DHGR) is an important yet difficult thing in the pattern recognizing groups and research communities.The latest establishment of new methods for acquisitions like Leap Motion and the Kinect permits deriving a vital data about hand pose which could be utilized for precisely recognizing the gestures.The previous system designed a Hidden Conditional Neural Field (HCNF) classifier with Leap Motion Controller (LMC) for DHGR.The descriptive information about the gestures by hand is obtained with the assistance of LMC.It keeps a note of the movement of hands and fingers in a digitalized manner and provides some points related to every gesture.Training and recognition is done with the help of this.However, HCNF has issue with overfitting problem due to the great expressive power, especially when trained on a small corpus.It reduces the classification accuracy.To solve this problem the proposed system designed a deep learning approach for DHGR.This system has 2 main phases such as feature extraction and classification.Initially, the acquired data from LMC is taken as an input and perform feature extraction from it.The given feature vector single-finger features and double-finger features.Based on extracted feature vector, Weighted Bias Mean based Convolutional Neural Network (WBMCNN) is utilized for DHGR.The demonstrated technique is analyzed on 2 sets of data of dynamic hand gesture accompanied by frames got with a Leap Motion Controller.The LeapMotion-Gesture3D recognizes with an accuracy of92.8% and Handicraft-Gesture dataset recognizes to an accuracy of 96.7%.This method can be specifically used for DHGR.The experimental result shows that it has a better capability in comparison with the previous methods in terms of accuracy and execution time.