Research On The Method Of Multi Person Behavior Recognition Based On Deep Learning

With the rapid development of science and technology,today's society is entering the era of big data.And the image data as the primary data carrier occupies an important position.In order to make full use of the image data,image recognition as a hot technology is rapidly developing,and the research on human behavior recognition has made some achievements.What's more,image recognition is widely used in the field of intelligent monitoring,motion analysis,etc.Existing methods related to human behavior recognition is mainly to identify the single behavior recognition instead of the study of multiplayer behavior recognition.However,it is more common and more important to study the multiplayer behavior recognition in practical applications.In order to make human behavior recognition to play a more important role in various fields,more and more scholars began to research on Multiplayer behavior recognition,and it can be found that Multiplayer behavior recognition compared with the simple single behavior recognition has many problems such as more characters which are difficult to distinguish,increasing the picture feature dimension which is difficult to learn,complex behavior background and more interference.To solve the above problems,this paper proposes a method of multiplayer behavior recognition based on convolutional neural network which has a significant effect in the feature learning applied to the multiplayer behavior recognition.Firstly,considering the complexity on Multiplayer behavior recognition,this paper selected double interactions as the research subject.What's more,it collected and pre-treated the image database required by experiments.Secondly,this paper extracted features from initial images by using unique and efficient dense-sift algorithm that behavior recognition isn't easily impacted by the shooting angle and light intensity.In addition,the resolution of feature images obtained by the algorithm isn't impacted by the size of initial images,which is conducive to the further calculation.Finally,this paper identified and categorized the feature images with LeNet-5 convolution neural network which had a good effect on the handwritten numeral recognition.Since the images of human behavior were more complex than those of handwritten number,this paper also changed the network input,the number of layers,the cores of filter,learning rate and the network output to improve recognition on human behavior.Furthermore,the experiments show that the proposed method termed Multiplayer behavior recognition has recognition ability of the boxing,hug and kissing of the double interactions.