| With the development of Internet and computer, smartphone is applied to moreand more areas. Acceleration sensor of smartphone is capable of measuirngacceleration of electronic devices. At the same time, with the development ofaccelerometer sensors, the applications of acceleration sensors are increasinglycommon in many areas of life. This thesis discusses the technologies of special actionrecognition based on embedded sensor of smartphone. Action recognition is a popularand important research topic in computer.According to main feature of activities with people indoor, walking and stoppingactions are regarded as normal condition, but falling activities are regarded asabnormal condition. In the thesis, the technology of special action recognition haswidely application prospect. The real-time falling detection can provide old peopletimely alert and call for help.Firstly, API with accelerometer sensor of Android is discussed. Subsequently, thesystem structure and configuration of action recognition is analyzed. The conception,main function, and the structure of algorithm are introduced, and also the popularalgorithm of KNN is analyzed. We abstract value of overweight, continuousweightless time, tilting angle, still time as features, and implement a decision treeclassifier running on an Android smartphone. The last section of the thesis analysesthe experiment and test of the action recognition system.The algorithm does not require fixing the oirentation of sensor We selected datafrom daily activities and a set of falling activities with similar acceleration feature fortest. The design of the algorithm system is completed. The results are that averageresponse-time of the algorithm is less than5seconds and average accuracy is morethan90%, which shows its effectiveness. Currently, research in the field of humanmotion recognition based on embedded sensor is still in its preliminary stage. Theresults of this thesis will promote the development of this technology to a certaindegree. |
PDF Full Text Request