Smartphone-based Human Activity Recognition Using Ubiquitous Positioning Data

Individual activities refer to the daily activities,such as home,work,study,travel,leisure and entertainment.The study of individual activities can provide basical data for many fields,including transportation planning,business,infectious disease prevention,etc.The acquisition of individual activity information traditionally relies on travel surveys.Large-volume travel surveys usually labor-intensive and time-consuming,which is difficult to meet the needs of social development.With the popular smartphones in recent years,built-in multiple sensors and powerful computing ability have enabled mobile phone to sense individual activities.On the other hand,indoor positioning technology has been rapidly developed.With smartphones,we can obtain indoor location through WiFi positioningCurrent studies on individual activities separately consider the indoor-and outdoor-activities.There is not a unified framework to acquire them.Based on the collection and processing of mobile phone sensor data,this thesis proposes a new method to automatically collect indoor-and outdoor-activities.The proposed method firstly develops one mobile phone app to collect indoor and outdoor positioning data.Outdoor positioning records is obtained by the GPS receiver module.The indoor positioning data is obtained by a K-neighbor-based WiFi positioning algorithm.Secondly,a set of moving time windows based positioning features are extracted,and the representative ensemble learning method are used to identify moving and stopping.Finally,the activity purpose is identified by machine learning.The spatial and temporal distribution characteristics of indoor and outdoor activities are analyzed.The findings of this study are:(1)The students mainly conducts study and scientific research activities in the study area.Indoors,it focuses on learning and communication,and rarely moves.This explain that students generally lack of exercise.(2)The activity chain of “dormitory ? learning ? canteen” is the most common,accounting for73%.This activity chains represents the activity mode of most students.(3)The learning activities is the lowest on Saturday in a week while the highest on Thursday.(4)Random forest has the best performance to recognizing the movement-stoppings and activity purpose.The overall movement-stoppings F1 value reaches 0.9132 and the overall activity purpose F1 value is 0.8775.After the trajectory point result is corrected by the filtering algorithm,the movement-stoppings F1 value is 0.9160.