Research On Dynamic Gesture Recognition Based On Multi-feature Fusion Algorithm

With the continuous breakthrough and innovation of high-tech,the traditional human-computer interaction methods that rely on keyboards,mice,handles,etc.have gradually developed into emerging interaction methods such as machine vision,pattern recognition,and artificial intelligence.In particular,the human-computer interaction method based on computer vision technology is gradually applied to all aspects of our production and life,and has become a hot research topic at home and abroad.Gestures,as the most intuitive and semantically diverse information communication method,have attracted the attention of researchers and become a popular research direction in the field of computer vision.However,with the in-depth research on dynamic gesture recognition,it is found that vision-based dynamic gesture recognition has many problems such as high background complexity,redundant information,fast change of gesture movements,and complicated changes in hand shape.Therefore,this thesis mainly focuses on the problems of the variability of dynamic gesture motion sequences and the complex spatial relationship of gestures,the redundant information in the gesture recognition process,and the recognition effect is easily affected by environmental changes and complex backgrounds in real scenes.The research work of this thesis is as follows:Aiming at the problems of high background complexity and redundant information of dynamic gesture recognition,this thesis firstly preprocesses dynamic gesture videos.Firstly,the key frame extraction algorithm based on frame difference method is used to extract key frames of dynamic gesture videos.On the basis of fully expressing the key content of gesture videos,the interference of redundant video frame information is avoided as much as possible.Next,in order to avoid the interference and influence of complex background on the recognition task,a 2D gesture model is established with the help of Open Pose pose estimation algorithm.11 skeletal joint points related to dynamic gesture actions are extracted,and a 2D gesture model is built for subsequent feature extraction tasks.It effectively avoids the interference of background,lighting,arms and palms blocked by clothing,etc.in the video,and further reduces the amount of model calculation.Aiming at the problems of changing gesture motion sequences and complex gesture space changes in dynamic gesture recognition,this thesis extracts the angle features and distance features of dynamic gestures as spatial feature descriptors,and extracts the trajectory features of dynamic gestures as temporal feature descriptors.Afterwards,the FV algorithm is used to encode the angle feature,distance feature and trajectory feature,and the weighted fusion mechanism is used for optimal weighted fusion.Finally,the recognition and classification of dynamic gestures are realized with the help of support vector machine.Through the weighted fusion of spatial features and temporal features,the gesture action can be described completely and accurately,which greatly improves the recognition accuracy of the algorithm.In order to prove the effectiveness of the algorithm proposed in this thesis,experiments are carried out on the UTD-MHAD dynamic gesture dataset and the Chinese traffic police command gesture dataset.Comparing and analyzing the excellent algorithms proposed in recent years,the experiments show that the algorithm proposed in this thesis has a certain improvement in the recognition rate,which verifies the effectiveness of the algorithm in this thesis.