Human Action Recognition Based On Intelligent Mobile Phone

The emergence of the smart phone is a very important milestone in the development of the mobile phone.Compared with the traditional function of mobile phone,smart phones can not only realize the basic function such as SMS,calls,more important is the smartphone has an independent operating system,can implement multitasking,wireless Internet access,strong expansibility,convenient and third-party software development,and various sensors to join the rich user experience and realize the application of more intelligent.With the addition of sensor devices,smart phones can be used in the safe driving,smartphone applications in human behavior recognition sentiment analysis,smartphone applications in medical care,health monitoring,smart phones,used for positioning and so on.Sensors have played a huge role,left the sensor cannot achieve these functions.In this paper,we study key technology for human action recognition based on smart phones,and introduces the traditional gesture recognition processing,respectively introduces the data preprocessing,feature extraction,classification method.Emphasis on the classification process of the algorithm used and compared the advantages and disadvantages of each algorithm,and the classification results under different scenarios.Considering the influence of the sensor position on classification,first identify the position sensor devices,and put forward a set of process,the emphasis on data feature extraction stage,combined with time domain and frequency domain characteristics in order to achieve better classification effect.Realized in MATLAB environment and experiments respectively,the experimental accuracy can reach above 90%.Developed online classified applications in Android platform,the experiment accuracy up to 85%.On the basis of the traditional decision tree classification algorithm,this paper puts forward a set of action recognition algorithm,is used to distinguish the stationary state,the stairs,down the stairs,walking,running state.Y axis acceleration signal cycle and amplitude,and the Z axis average size classification as the basis.To obtain the signal cycle,this paper proposes a new algorithm for approximate periodic signal,Different from conventional methods for signal cycle,the traditional method is to use a function first derivative and second derivative.In this paper,algorithm based on function monotonicity principle and concept of extreme obtained approximate cycle.Experimental proof enough to discern the upstairs,walk and go downstairs.The highest accuracy can reach 88%.And compared under differentprocessing window size classification accuracy,finding the right window value is to make the highest classification accuracy is some problems which need to be studied further in the future.