Research On A Human Behavior Recognition Algorithm Based On N-SCHMM

With the rapid development of information technology and computer hardware and software technology,it is unnecessary and difficult to only rely on human beings to process the massive information and data.It needs intelligent equipment like computer or machine instead of human eyes to understand the behavior of people and their expression of the semantic information.There are several problems in the research of human behavior recognition.Video-based human behavior recognition usually contains many frames,how to extract the typical key frames that can express the behavior,and thus improve the efficiency of the recognition behavior while reducing the amount of data;Markov model of human behavior recognition usually use a fully connected hidden Markov model,resulting in a lot of redundant state and a lot of calculation,how to improve the efficiency of human behavior and reduce the computational complexity by applying hidden Markov model.In view of the above problems,this thesis studies the method of recognizing the human behavior based on video using the hidden Markov model.The main contributions are summarized as follows:(1)Proposed a secondary sampling time domain detection algorithm to determine the key frame.In a simple behavior of the video data,human behavior can be seen as a few typical key frames,and other frames can be seen as the transition between these key frames.The binary image sequence of the moving body is obtained,the contour of the human body is extracted,the coordinate system is established based on the human centroid.Correlation coefficient is used to determine the inter-frame similarity because the human behavior is cyclical.Based on the process of periodic sampling frame,the key frame may be missed,so the use of secondary sampling time domain detection can reduce the sampling time window and increase the key frame check rate.(2)Proposed a MIK-means clustering algorithm to extract key frames.The shortcoming of the K-means clustering algorithm is that selection of several initial clustering centers is usually a stochastic process,and the result of the final clustering is more dependent on the selection of several initial clustering centers.Therefore,The clustering center of several clusters is clustered as the key frame sequence of the behavior cluster,and the clustering center of the cluster is selected as the initial clustering center.Extracting the key frames in the video can effectively reduce the amount of computation in the recognition process and improve the efficiency of recognition.(3)Proposed an improved N-SCHMM model to identify the behavior of human body.Because the human behavior is continuous,it is consistent with the assumption that the current state is only related to its previous state.Hidden Markov model is commonly used when identifying human behavior is a fully connected model,without taking into account the relationship between the behavioral characteristics of human behavior and the hidden Markov topology.In this paper,we propose an improved N-SCHMM model to design the topology according to the motion characteristics of the human body,limit the condition of state transition,and reduce the number of redundant states.We propose to introduce the weight in the state output probability training of N-SCHMM model.Considering the continuity of human behavior,we introduce the weight of state output probability,make the training parameters more in line with the movement characteristics of human behavior,reduce the randomness of state output,reduce the complexity of operation and improve the efficiency of recognition behavior.