Study On Smart Monitoring Based On Motion Sensor Environment

At present, China is in the stage of aging society. As the children often work farfrom home, the number of Empty Nest Elder is rising steadily, and the demand of smartmonitoring system is urgent. To solve the problem, it is essential to recognize theactivities of the aged. For now, there are two kinds of approaches to recognize activities:video-based and sensor-based. Video-based approaches have well proven techniques,but it is not capable for home settings because it is an invasion of privacy during dataacquisition. With the development of wireless sensor network, it attracts more and moreattentions of the researchers to study on activity recognition based on wireless sensornetwork.In this thesis, human activity recognition is combined with the wireless sensornetwork. To solve the problems in smart monitoring of home settings, the distributedabnormal activity detection approach which is called DetectingAct is proposed. Becausethe definition of abnormal activity is subjective, the normal activity is defined as therepetitive activity in data and the abnormal activity is defined as the activity whosedeviation between normal activities is big enough.The main contributions of this thesis are shown as follows:Firstly, to meet the real-time demand of smart monitoring, distributed abnormalactivity detection approach DetectingAct is proposed. DetectingAct, which employs thelimited computing and storage resources of these sensor nodes, avoids the bandwidthproblem and response time problem of centralized detection method, and improves thedetection speed and accuracy.Secondly, according to the defects in the definition of activity in traditionalresearches, the activity model with duration information is proposed. The modifiedactivity model provides more accurate definition of activity and improves the detectionaccuracy.Thirdly, due to the lack of simulation system of motion sensor, the simulation andstatistical system for ADL is designed. The system is capable to simulate the smartenvironments in real life with high credibility of simulation data.Finally, tests under the simulation system and real environments are conducted toprove the accuracy, real-time property and stability of distributed abnormal activitydetection method DetectingAct compared with the traditional trajectory-based detection method.