CITATION:
Aloba, A. 2021. Tailoring Motion Recognition Systems to Children’s Motions. Ph.D. thesis, Department of Computer and Information Science and Engineering, University of Florida. August 2021. [PDF]
ABSTRACT:
“Motion-based applications allow users to interact with a system or interface using motions. These applications often require accurate recognition of motions to ensure meaningful interactive experiences and are becoming increasingly popular among children. However, motion recognition systems are usually trained on adults’ motions even though children move differently from adults. Our findings showed that na¨ıve viewers could perceive the difference between children’s and adults’ motions at levels significantly above chance when the motion was abstracted from appearance cues (e.g., height). Our findings further showed that skeleton-based motion recognition systems, which accept as input the positions of joints tracked by a motion sensor in 3D space over time, perform poorly on children’s motions compared to adults’ motions. Therefore, motion recognition systems should be tailored to children’s motion qualities to enable accurate recognition.
To characterize children’s natural motion qualities, we focused on understanding how children perform motions. We designed a method to analyze the key body parts that users move during a motion and found that children perform motions more inconsistently compared to adults. To understand why children’s motions are more inconsistent, we quantified the differences between children’s and adults’ motions. We initially analyzed children’s and adults’ walking and running motions using gait features and found that children move in less time and with higher energy. Then, we generalized this analysis to a broader set of motions by defining 24 articulation features (11 of which we newly identified), that quantitatively describe users’ motions performance. We analyzed these features on a subset of children’s and adults’ motions from the Kinder-Gator dataset to reveal new insights about how children perform motions (i.e., their motion qualities).
Based on our findings, we propose guidelines for tailoring motion recognition systems to children’s motion qualities to enable accurate recognition of children’s motions and guidelines for designing motion sets and motion applications for children. The implementation of these guidelines to tailor actual recognition systems to children’s motions and evaluate their performance is out of the scope of this dissertation work. Finally, we also provide ideas for continued research in improving children’s interactive experiences in motion-based applications.”
Files:
Funding:
This work is partially supported by National Science Foundation Grant Awards #IIS–1218395 / IIS–1433228 and IIS–1552598. Opinions, findings, and conclusions or recommendations expressed in this paper are those of the authors and do not necessarily reflect these agencies’ views.