Research On Activity Recognition Technology Based On Noninvasive Sensors

In recent years,more and more elderly people cannot live by themselves in their homes,and the emergence of smart home system can solve this problem to some extent.Smart home system can bring residents a more comfortable living environment by recognizing daily activities automatically and help to improve elderly daily life.Therefore,accurate activity recognition is very significant in a smart home environment.Although technology research of activity recognition has made significant progress in a smart home environment,activity recognition still presents many challenges in a smart home.First,the existing sensor segmentation algorithms usually divide sensor data into too many fragments,which cannot guarantee the integrity of users’ activity during activity detection.Second,some existing classification algorithms have the disadvantage of low accuracy of activity recognition.Finally,there may be abnormalities in users’ daily life activity in a smart home,and most researchers only focus on normal activity patterns,ignoring the expression of abnormal activity.Based on the above problems,this paper aims to recognize activity in a smart home based on non-invasive sensors and conduct the following three researches:(1)A segmentation algorithm based on predefined knowledge is proposed to segment the raw sensor data to get the sensor fragments,and then the fragments are merged into a complete activity fragment using the fragment merging algorithm based on the topic model.Compared with the existing methods,the proposed method has lower segmentation error rate and fragment ratio.(2)A modified K-Nearest Neighbor(KNN)algorithm combined with center distance is proposed to recognize the activities of sensor fragments.Compared with several existing baseline classifiers,the modified KNN algorithm has better accuracy of activity recognition.(3)A density-based clustering algorithm is used to analyze the same activity,to compare the inherent differences between the activities,and to separate the anomalies in the same activity at different time.The experimental results show that the abnormal activity detection method in this paper can detect two kinds of abnormal activity,and can achieve the purpose of monitoring users’ health status.