Abstract | ||
---|---|---|
Physical activity recognition is an active area in wearable computing community. One challenge is to handle large amount of data which does not belongs to pre-defined classes. In this paper, we propose an unsupervised approach to discover frequent Activities of Daily Living (ADL) patterns from wearable acceleration data streams. The approach discovers frequently-occurred ADL patterns and learns a classifier for each discovered pattern. The number of ADL patterns is dynamically decided at runtime. The proposed approach can be divided into two phases: A rough segmentation (pattern discovery) of the motion stream via topic model, and classifier learning with semi-supervised learning, whose training set is obtained by segmentation results from the first stage. Experiments on real life datasets show promising results on both ability to discover ADL patterns and recognizing accuracy of its learned classifiers. |
Year | DOI | Venue |
---|---|---|
2014 | 10.1145/2670473.2670483 | VRCAI |
Keywords | Field | DocType |
adl discovery,pattern discovery,topic model | ADL patterns,Data stream mining,Activity recognition,Segmentation,Accelerometer,Computer science,Wearable computer,Artificial intelligence,Topic model,Classifier (linguistics),Machine learning | Conference |
Citations | PageRank | References |
0 | 0.34 | 19 |
Authors | ||
2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Zehong Tan | 1 | 0 | 0.34 |
Shuangjiu Xiao | 2 | 41 | 14.18 |