Research And Application Of Real Time Dynamic Gesture Recognition Technology

With the continuous development of artificial-intelligence interaction technology,the traditional interaction system based on keyboard and mouse has been gradually replaced by natural AI interaction.Gesture interaction emerged as one of the mainstream ways of new natural interaction and it has wide application prospect.Due to the complex diversity,one of the key natural interaction technologies is to study how to carry out real-time gesture recognition.This paper focuses on the feature extraction and recognition of real-time dynamic gesture interaction technology.The main work is as follows:(1)The paper proposes a dynamic gesture feature extraction algorithm based on key frames and local extremum(KFLE),as the solution of intensive computation and long calculation of the existing algorithm under the gesture interactive environment of computer vision.Firstly,paper implements a key frame gesture image selection algorithm based on motion direction and adaptive threshold,and select key frame gesture image with obvious change of direction or distance in dynamic gesture sequence.During feature extraction processing,we put forward a fingertip feature extraction algorithm according to partial extremum.Then paper specifies the fingertip feature point by finding the local extreme point using the distance function and convex hull filtering.The experimental results show that the KFLE algorithm can shorten the feature extraction time by 44.3% and 71.9% respectively,compared with the gesture feature extraction algorithm based on convex hull defect and the gesture feature extraction algorithm based on improved k-curve.(2)The dynamic gesture recognition algorithm based on DTW is suitable for flexible and complex environment because of its simple calculation and no training.However,the algorithm originates from dynamic programming,therefore has high computational complexity and long recognition time.Paper raises a real-time dynamic Gesture Recognition Algorithm based on Boundary-Constraint DTW(BCDTW).First,a method of converting a two-dimensional gesture sequence of unequal length into an equal length one-dimensional gesture sequence is given.Then BCDTW reduces the calculation between the gesture sequences,which do not satisfy the similarity requirement,by obtaining the lower bound of the DTW distance.Hereafter BCDTW determines the gesture border width,sets the matching gesture data range.The experimental results show that the BCDTW algorithm can shorten the gesture recognition time by 50.4% compared with the algorithm based on DTW.(3)Finally,we implement the gesture interactive prototype system thanks to above research,aiming to utilize the urban air quality data visual query system through gestures.The prototype system includes gesture operation management,custom gesture and gesture interaction,and verifies the validity and practicability of KFLE algorithm and BCDTW algorithm.