Improving activity recognition without sensor data

Abstract

Wearable sensing systems, through their proximity with their user, can be used to automatically infer the wearer's activity to obtain detailed information on availability, behavioural patterns and health. For this purpose, classifiers need to be designed and evaluated with sufficient training data from these sensors and from a representative set of users, which requires starting this procedure from scratch for every new sensing system and set of activities. To alleviate this procedure and optimize classification performance, the use of time use surveys has been suggested: These large databases contain typically several days worth of detailed activity information from a large population of hundreds of thousands of participants. This paper uses a strategy first suggested by [16] that utilizes time use diaries in an activity recognition method. We offer a comparison of the aforementioned North-American data with a large European database, showing that although there are several cultural differences, certain important features are shared between both regions. By cross-validating across the 5160 households in this new data with activity episodes of 13798 individuals, especially distinctive features turn out to be time and participant's location. Additionally, we identify for 11 different activities which features are most suited to be used for later on activity recognition.

Prior knowledge of human activities from social data

“…[6] explore the use of temporal features for classification and [5] shows significant improvement of recognition performance using a daily activity rhythm model. Recently, work of Partridge et al [2], extended by [1], propose the use of government-conducted time-use surveys to generate prior knowledge from thousands of subjects. Although wellannotated, the data is costly to produce and as a result, not freely available for all countries.…”

“…In order to allow AR systems to directly access prior information from crowd-generated activity reports, we need to provide structure to quantify the odds of activity priors. Similar to [2,1], we make use of the 2011 American Time-Use Study (ATUS) taxonomy [4] to categorize our raw activities data. We label "tip" texts into 17 tier-1 activity categories according to [4] via Amazon Mechanical Turk.…”

Using time use with mobile sensor data

“…Recently, researchers in [4] investigated the GTUS in regard to benefits for activity recognition, identifying features in the dataset that can be used to determine the activities that occurred within the time use dataset. They extracted activity histograms according to the investigated features, e.g., time or location.…”

“…Similar to [4], the work in [16] investigates the American Time Use Survey (ATUS) and identifies time use surveys as a promising instrument for designing activity recognition systems. The ATUS can be freely obtained online 2 for further studies, while the GTUS is available for regional government employees only.…”

“…Time use information is already available through national time use surveys, which are statistical surveys that contain information from thousands of individuals on where and when people are performing which activities on average. Recent work [4] used such data bases to infer the user's activity based on the activities of similar users within the survey. Results have thus far shown that time survey data, when used from within the same geographical and cultural region as the user, could enhance activity recognition systems considerably.…”