Font Size: a A A

Research And Application Of Gesture Recognition Based On Deep Learning

Posted on:2022-04-17Degree:MasterType:Thesis
Country:ChinaCandidate:Z Y GuoFull Text:PDF
GTID:2518306326484704Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
With the rapid development of artificial intelligence technology and computer equipment,human-computer interaction technology has become an indispensable part of today's life.Gestures are simple and flexible,making gesture recognition a popular research direction in human-computer interaction technology.Traditional gesture recognition is based on manually extracting features,which has low accuracy and poor robustness.In recent years,deep learning has made great achievements in computer vision fields such as image classification and target detection,providing a new direction for the research of gesture recognition.This paper mainly studies gesture recognition based on deep learning.The main research work includes the following points:1)Research on gesture recognition using deep learning network model.Two deep learning network models,Faster R-CNN and YOLOv4,are used to recognize gestures respectively.By comparing the standard data set and the self-made data set,the two recognition results are analyzed,it is found that the recognition results of YOLOv4 algorithm are higher than the Faster R-CNN algorithm in confidence,target frame positioning accuracy,and detection time.It is faster,but there are still problems such as inaccurate positioning of the gesture target frame,low recognition confidence,and low recognition accuracy of small target gestures,so the YOLOv4 algorithm needs to be improved.2)Improve the YOLOv4 algorithm.Aiming at the problems of traditional YOLOv4 recognizing gestures,the K-means clustering algorithm is used to re-adjust the size of the prior frame to improve the accuracy of the position of the prior frame.By optimizing the network structure,the resolution of the original feature map is changed to improve the recognition of gestures.Accuracy and efficiency.Experimental results show that the confidence and recognition accuracy of gestures have been greatly improved,small target gestures can be accurately recognized,and the goal of real-time gesture recognition can be achieved.3)Use the improved YOLOv4 algorithm to build a gesture recognition application system.The gesture recognition based on the improved YOLOv4 algorithm will be applied to simple human-computer interaction.Build a three-dimensional scene and use the improved YOLOv4 network to recognize different gestures in real time,thereby controlling the tank model in the scene to perform corresponding actions.
Keywords/Search Tags:gesture recognition, gesture interaction, target detection, deep learning, YOLOv4
PDF Full Text Request
Related items