Research On The Recognition And Understanding Of The Human Interaction Behavior Based On Video

Human interaction recognition is an important research topic in computer vision direction,which has broad application prospect in the fields of video retrieval,intelligent home and smart surveillance system,etc.However,there are two difficult problems in this study.The first one is the problem of the recognition framework,i.e.the traditional recognition framework is hard to be applied in this study.Another difficulty is how to find more appropriate feature representation methods to describe inner relations between individuals of interaction.At present,there are two interaction recognition frameworks,i.e.the method based on the holistic recognition and the method based on the individual segmentation.The inherent problems of existing recognition methods are studies in depth and shown below:Firstly,in the holistic recognition framework,the problem of balance between computational complexity and recognition accuracy still exists in human interaction recognition.To solve this issue,an innovative way based on statistical features of key frame feature library was proposed.It established the key frame feature library by the “Bag of Words(BOW)” and represented interactive videos statistically.The primary advantages of this approach are that the feature is simple,no modeling is required with good recognition results,and the real-time processing can be basically realized.Secondly,as the human interaction recognition methods based on motion co-occurrence had relatively low recognition accuracy,an innovative and effective method based on co-occurring visual matrix sequence was also proposed.This method sufficiently combines the advantages of the co-occurring visual matrix and probability graph model,which not only enriches the hidden information of the videos but also improves the discriminative ability of different interaction behaviors.The method achieves better recognition performance with simple implementation.Finally,according to the problem existed in recognition framework,a human interaction recognition method based on multi-stage framework and co-occurring visual matrix sequence was proposed to increase interaction diversity.This method can describe the intrinsic properties of interaction effectively,and avoid interference such as overlap.The method has high recognition accuracy and can satisfy real-time requirement.