Inexact Daily Activity Feature Extraction Method In Smart Home

The research goal of daily activity recognition in smart homes is to infer what kind of daily activities residents are performing by analyzing environmental data sensed by sensors deployed in smart homes.Daily activity recognition is the basis for realizing smart home.The smart home system can provide residents with health protection according to the daily activity recognition results,effectively save nursing costs,and also provide personalized help and services for residents' daily life.The effect of daily activity recognition depends on the quality of sensor event segmentation,but sensor event segmentation not only requires the study of specific algorithms,which increases development costs,but also requires a complete cycle of stream of sensor events during the algorithm training process,so it is not suitable for timely and specific moment daily activity recognition.In order to improve the timeliness of daily activity recognition,this paper proposes an approach for timely daily activity recognition from the sensor event stream triggered first.This method extracts the time boundary of beginning sensor event and the beginning sensor set from sensor event stream triggered first by a daily activity as the daily activity feature space.Then use the proposed timely daily activity recognition algorithm to calculate the similarity of the sensor event stream triggered first in the training set and the test set,and determine the types of daily activity based on the similarity.In addition,this paper also proposes daily activity recognition for specific moments.The method first extracts the boundary sensor event of the daily activity instance from the training set as the daily activity feature space according to the specified moment.Then determines the approximate boundary sensor event of the daily activity at the specified moment in the test set according to the boundary sensor event in the training set.Finally,daily behavior recognition is performed by calculating the similarity between the sensor event streams of the training set and the test set.The method proposed in this paper is verified on the public CASAS daily activity identification data set.The optimal Recall and Precision of timely daily activity recognition based on the sensor event stream triggered first have reached 91.0%and 91.7%,respectively.The lowest saved time proportion is 41.39%.The optimal Accuracy,Precision and F-measure of daily activity recognition for specific moments have reached 87.99%,90.78%and 88.45%respectively.The experimental results verify the effectiveness of the proposed method in this paper.