Research On Human Movement Recognition Method Based On Multi-feature

With the continuous development of computer technology,human motion recognition technology has become an indispensable research content in the field of computer vision.It plays a huge role in intelligent analysis and other intelligent needs in intelligent video monitoring,intelligent monitoring,video content analysis and other fields.This paper studies the method of body movement recognition.The specific work is as follows:1.To solve the problem of low recognition rate caused by many impractical optical flow points in the inter-frame optical flow method due to transparency,noise,multiple light sources and occlusion,this paper proposes a recognition method based on the combination of three-dimensional space-time interest points and optical flow detection.When three-dimensional spatio-temporal interest points are detected,Gaussian filter is used to filter the spatial domain,and two orthogonal one-dimensional Gabor filters are used to filter the time domain.After the detection area is obtained,the optical flow diagram is obtained by optical flow method,and then the classification and recognition is performed after feature extraction.Experimental results show that this method has a higher recognition rate for KTH single person behavior video database.2.An improved LBP feature is proposed in this paper,which only considers the relationship between the central pixel and several neighboring pixels in a cell,but ignores the role of the central pixel and the relationship between neighboring pixels.By introducing the gray value difference between the center pixel and the neighborhood pixel and taking the mean value as the threshold value,the difference between the center pixel and the field pixel is compared with the threshold value,and a new coding method is adopted to give new features.For Weizmann and You Tube human behavior databases,experimental results show that this method has a higher recognition rate than LBP feature method.3.Because the HOG descriptor generation process is lengthy and the feature extraction efficiency is slow,the real-time performance is greatly reduced.Meanwhile,the noise sensitivity of HOG feature and LBP feature leads to a low recognition rate.This paper proposes a multi-feature fusion human motion recognition method.Experimental results of Weizmann and You Tube human behavior databases show that this method has high recognition performance.