With the recent development of cheap and accurate depth sensors, activity recognition research has largely focused on the use of features created from 3D, rather than 2D, skeletal joint location. Nevertheless, conventional 2D RGB cameras remain an attractive data collection tool due to their low cost and ease of use. This study investigates the benefits of using 2D skeletal joints for activity recognition using visualisation and an exemplar classifier. Results show that 2D models can be as informative as 3D models, demonstrating the informativeness of joints extracted from RGB video.
|Title of host publication||Irish Machine Vision & Image Processing Conference proceedings|
|Subtitle of host publication||IMVIP 2019|
|Publisher||Irish Pattern Recognition and Classification Society|
|Number of pages||20|
|ISBN (Electronic)||ISBN 978-0-9934207-4-0|
|Publication status||Published - Aug 2019|
|Event||Irish Machine Vision & Image Processing IMVIP 2019 - Grangegorman Campus, Technological University, Dublin, Ireland|
Duration: 28 Aug 2019 → 30 Aug 2019
|Conference||Irish Machine Vision & Image Processing IMVIP 2019|
|Period||28/08/19 → 30/08/19|
- Activity Recognition
- 3D Skeletal Joints
- 2D Skeletal Joints
Marshall, F., Zhang, S., & Scotney, B. (2019). Comparison of Activity Recognition using 2D and 3D Skeletal Joint Data. In Irish Machine Vision & Image Processing Conference proceedings: IMVIP 2019 (pp. 13). Irish Pattern Recognition and Classification Society.