Activity Analysis Algrithom Based On Hierarchical Cognition Model

With the rapid development of information society and intelligent society,the information processing ability of people can not meet the data acquisition ability of equipment.So the artificial intelligence techniques are researched to help people analyse information and make decision.Computer vision technology is an important part in artificial intelligence.Activity analysis technology and an important research field of computer vision.By analyzing the image of characters collected by the camera such as the postures and movements,computer can automatically judge what kind of activity the characters are doing.The research can be applied in the field of public safety,finance,medicine,automatic driving,robot,etc.This paper mainly discusses the key problems in the field of activity analysis and proposes solutions to these key problems.By studying the history of the development of computer vision model,this paper presents a hierarchical cognitive model which fits the purpose visual model to solve the problem of activity analysis.The model divides the activity analysis into three levels: the feature layer,the abstract layer and the cognitive layer.The feature layer includes activity feature perception problem.The abstraction layer includes activity encoding problem.The cognitive level includes the decision analysis of activity.In this paper,the role of each layer in activity analysis,the difficulties of its own research and the solutions to problems are discussed.Four kinds of activity analysis algorithms which focus on different levels are put forward.These algorithms make theoretical innovation and construct activity analysis model at different levels.It solves the key problems of activity analysis in different levels and improves the calculation theory in the field of activity analysis.It also improves the active perception ability of activity analysis model.The ability of automatic learning improves the accuracy of activity analysis.The main contributions of the paper are listed as follows:1)In this paper,a hierarchical cognitive model is proposed to solve the problem of activity analysis.The problem is divided into three levels:feature layer,abstract layer and cognitive layer.The key problems,research difficulties and solutions in each layer are elaborated in detail.2)In this paper,an activity analysis algorithm based on two-stream dictionary learning structure is proposed.The algorithm focuses on the theoretical innovation of the feature layer and the abstract layer in hierarchical cognitive model.We proposes a feature to describe the activity of RGB video.This feature has good scale invariance and rotation invariance.It improves the performance of activity analysis.3)In this paper,an activity analysis algorithm based on hierarchical sparse coding and intra block segmentation is proposed.The algorithm focuses on the theoretical innovation of the feature layer and abstract layer in the hierarchical cognitive model.A hierarchical sparse coding structure is proposed,which reduces the feature redundancy,reduces the loss of coding information and improves the abstraction of the coding.The accuracy of activity analysis is improved.4)In this paper,an activity analysis algorithm based on adaptive classification network and intra frame split strategy is proposed,which focuses on the theoretical innovation of the abstract layer and the cognitive layer in the hierarchical cognitive model.The adaptive classification network is constructed by using deep learning technology.The network has good automatic perception ability and automatic learning ability.Its structure has intra frame adaptability,which is suitable for abnormal behavior detection tasks and improves the accuracy of activity analysis.5)In this paper,an activity analysis algorithm based on temporal sparse auto-coding network is proposed.The algorithm focuses on the theoretical innovation of the feature layer,abstract layer and cognitive layer in the hierarchical cognitive model.The network is constructed by using deep learning technology,and the activity features are encoded.The network has good automatic cognition ability and automatic learning ability.It improves the similarity of the same category activity and improve the distinguishability of different types of activity.It reduces the redundancy of activity characteristics,and improve the accuracy of activity analysis.