Name
Abstract | ||
|---|---|---|
Active games augment physical activity with the immersive elements of computer games running on mobile devices to encourage healthy behaviours in players. Such games rely on automated human activity recognition systems to obtain meaningful input for gameplay. The nature, number and placement of the sensors used to measure game activities affects the quality of activity recognition and the resulting value that this has for game play. This study investigates the recognition performance impact of using parallel sensing strategies from multiple body locations, as well as multiple sensor modalities without data fusion. C4.5 decision trees are trained on both raw sensor data and extracted features and classification performance is evaluated with tenfold cross-validation and 80/20 training/test methods. It was found that recognition accuracy depends on location and sensor types. Best results are achieved at locations closer to the core of the body. Classifiers derived from other sensor data achieve comparable performance to triaxial accelerometers. This study suggests that exergame hardware would benefit from incorporating multiple sensor modalities into a single device. |
Year | DOI | Venue |
|---|---|---|
2017 | 10.1145/3014812.3014875 | ACSW |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
6 | 4 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Natalie Jablonsky | 1 | 0 | 0.68 |
| Sophie McKenzie | 2 | 0 | 0.68 |
| Shaun Bangay | 3 | 97 | 17.72 |
| Tim Wilkin | 4 | 0 | 0.34 |