Research Of Visual Event Modeling

Event detection has drawn much attention in the field of computer vision,intelligent monitoring,social networks,human-computer interaction and many other fields.Nowadays,how to use the computer effectively to complete the task of event detection in huge amounts of data is still a very challenging problem for researchers.Common event detection method often involves the technique computer vision,image processing,machine learning,pattern recognition and data mining.Based on the investigation of previous research,my work focus on feature representation and event detection modeling method,whose effectiveness can be verified by the experiment on human action recognition and mitosis event detection.For human action recognition,we realize a cross-view action recognition method based on transferable dictionary pairs(TDP).After the feature extraction of action video,feature captured in the source view is applied to source view by sparse coding.Forcing the same action videos captured from two different views have the same sparse representation,the transferable dictionary bridges the gap of features between the two views.Finally,k-NN algorithm classifies the feature transferred.As for mitosis event detection,we propose a mitosis event detection framework based on slow feature analysis(SFA).Traditional slow feature analysis strategy is extended to three kinds of learning strategies,producing distinctive slow feature function depending on different input data.In mitosis cell classification process,we apply support vector machine(SVM)and hidden conditional random field(HCRF)to find the cells that most likely contain mitosis event.Results proves that dynamic model has better performance.