Application Of Clustering Algorithm Based On Validity Index In Analyzing The Behavior Feature Data Of Same Behavior From Multi-View

Now the demand for convenient and intelligent life is increasing. The human behavior recognition and analysis that based on visual have been widely applied in the field of human-computer interaction, safety, education and teaching. Therefore, it has a great value and practical significance to identify and analyze the behavior with high accuracy. However, the ambiguity, relevance and diversity of behavior lead to the unpredictable structure of behavior feature data and the difficulty to judge the effect of clustering when do the clustering analysis of behavior feature data. In this paper, we analyze and research the clustering validity with different clustering algorithms, under the background of the multi-view behavior recognition, and apply different algorithms with validity index into the analysis and modeling of the multi-view behavior feature data, and they achieve good results.This paper elaborates and analyzes the clustering evaluation principle in the situation that the division is unknown, and also introduces the common validity indexes and refers to the advantages of each of them, proposes a validity index (CS) which is based on the inter-cluster compactness and separation of clusters. We combine the validity index with FCM, GK, GG algorithms respectively, and compare it with other indexes through experiments to verify the validity of the index.We extract Spatial -Temporal Interesting Points from the video of IXMAS multi-view video dataset, and build recurrence plot for the same behavior under five views to get the multi-view behavior feature data, then cluster them to get the behaviors recognition model, and use it as the testing template to test a variety of behavior and judge if they are same to the behavior that the template represents or not. And the clustering effect plays an important role in the final behavior recognition. Good clustering is helpful to build a good recognition model, which improves the recognition rate. In this paper, we apply different clustering algorithms into analyzing the multi-view behavior feature data, and these algorithms are based on clustering validity index, combine the index with FCM, GK and GG clustering algorithms respectively, and analyze the behavior feature data of walk, kick and sit down respectively from five views. According to the evaluation of the validity index and the verification results of multi testing video segments, the index can determine the optimum cluster number effectively, and provide a good foundation for build the behavior template which depends on the clustering results.