Using Neural Networks to Distinguish Children’s Age with Visual Features of Sketches

Aniket Patel; Seth Polsley; Tracy Anne Hammond

Download Paper | Permalink

Conference: 2022 ASEE Gulf Southwest Annual Conference
Location: Prairie View, Texas
Publication Date: March 16, 2022
Start Date: March 16, 2022
End Date: March 18, 2022
Page Count: 7
DOI: 10.18260/1-2--39220
Permanent URL: https://peer.asee.org/39220
Download Count: 368

Paper Authors

biography

Aniket Patel Texas A&M University

visit author page

Aniket Patel is a junior in Computer Science at Texas A&M University. He is working as an undergraduate researcher pursuing how children’s drawing ability links to other developmental features associated with learning and how machine learning can be applied to this space. He previously worked as a researcher studying material science and analyzed material diffraction patterns.

visit author page

biography

Seth Polsley Texas A&M University

visit author page

Seth Polsley is a PhD student at Texas A&M University in the Sketch Recognition Lab under Director Tracy Hammond. His research interests may be broadly classified as “intelligent systems," with an emphasis on studying and building interactions that merge the capabilities of computers with the intuitive behaviors of humans. He holds a Masters and Bachelors in Computer Engineering from Texas A&M and University of Kansas, respectively, and has previously worked at Lexmark International and MIT Lincoln Lab.

visit author page

biography

Tracy Anne Hammond Texas A&M University orcid.org/0000-0001-7272-0507

visit author page

Dr. Hammond is Director of the Texas A&M University Institute for Engineering Education & Innovation and also the chair of the Engineering Education Faculty. She is also Director of the Sketch Recognition Lab and Professor in the Department of Computer Science & Engineering. She is a member of the Center for Population and Aging, the Center for Remote Health Technologies & Systems as well as the Institute for Data Science. Hammond is a PI for over 13 million in funded research, from NSF, DARPA, Google, Microsoft, and others. Hammond holds a Ph.D. in Computer Science and FTO (Finance Technology Option) from the Massachusetts Institute of Technology, and four degrees from Columbia University: an M.S in Anthropology, an M.S. in Computer Science, a B.A. in Mathematics, and a B.S. in Applied Mathematics and Physics. Hammond advised 17 UG theses, 29 MS theses, and 10 Ph.D. dissertations. Hammond is the 2020 recipient of the TEES Faculty Fellows Award and the 2011 recipient of the Charles H. Barclay, Jr. '45 Faculty Fellow Award. Hammond has been featured on the Discovery Channel and other news sources. Hammond is dedicated to diversity and equity, which is reflected in her publications, research, teaching, service, and mentoring. More at http://srl.tamu.edu and http://ieei.tamu.edu.

visit author page

Download Paper | Permalink

Abstract

Children’s fine motor control is linked to other critical skills during early childhood development, including school-readiness, reading comprehension, and math performance. Detecting fine motor delay is important to ensure healthy development and avoid delays in other areas.

Experts use a variety of methods to assess fine motor control, including drawing ability, which is closely linked to fine motor ability. Most assessments that use sketching rely on shape correctness. However, this requires a human expert to evaluate the drawings and uses only visual features. Furthermore, due to lack of awareness, costs associated with assessments, and access issues, parents may not know that their child is developmentally delayed until additional therapy is needed.

We seek to provide parents and teachers with easier access to fine motor skill assessment. By implementing sketch recognition and machine learning capabilities into tablet devices, we envision children drawing and playing on tablets while also empowering their parents to track their motor development and be alerted to potential issues.

In this work, we develop a vision-based approach using neural networks to distinguish childrens’ age groups by considering only their drawings. We use a dataset of 551 sketches from 57 children aged 3 to 8 years old to detect if they are 4 or under versus 5 or older. This effort is part of a larger project which uses sketch recognition to assess children’s fine motor ability. We focus on vision-based assessment to distinguish approximate ages in order to compare with human expert evaluation on the same images. By capturing the vision-based components of the sketches, similar to what a human evaluator may use, we hope to enable software that performs similarly to expert level.

Our neural network achieved generalization and validation accuracies of 75.0% and 66.7%, respectively, with a set of curve drawings, and a generalization and validation accuracy of 72.3% with the corner dataset. By comparison, we also asked two human evaluators to label age groups based on a portion of the dataset, and they achieved 77.5% and 66.7% for the curve and corner drawings, respectively. In general, the human judges struggled more with corner drawings, and these findings provide support for future work integrating more sketch recognition features, such as pressure and timing information, in order to make better assessments that are fast, easy, and accessible for parents and children.

Citation
Format

Patel, A., & Polsley, S., & Hammond, T. A. (2022, March), Using Neural Networks to Distinguish Children’s Age with Visual Features of Sketches Paper presented at 2022 ASEE Gulf Southwest Annual Conference, Prairie View, Texas. 10.18260/1-2--39220

TY - CPAPER
AB - Children’s fine motor control is linked to other critical skills during early childhood development, including school-readiness, reading comprehension, and math performance. Detecting fine motor delay is important to ensure healthy development and avoid delays in other areas.

Experts use a variety of methods to assess fine motor control, including drawing ability, which is closely linked to fine motor ability. Most assessments that use sketching rely on shape correctness. However, this requires a human expert to evaluate the drawings and uses only visual features. Furthermore, due to lack of awareness, costs associated with assessments, and access issues, parents may not know that their child is developmentally delayed until additional therapy is needed.

We seek to provide parents and teachers with easier access to fine motor skill assessment. By implementing sketch recognition and machine learning capabilities into tablet devices, we envision children drawing and playing on tablets while also empowering their parents to track their motor development and be alerted to potential issues.

In this work, we develop a vision-based approach using neural networks to distinguish childrens’ age groups by considering only their drawings. We use a dataset of 551 sketches from 57 children aged 3 to 8 years old to detect if they are 4 or under versus 5 or older. This effort is part of a larger project which uses sketch recognition to assess children’s fine motor ability. We focus on vision-based assessment to distinguish approximate ages in order to compare with human expert evaluation on the same images. By capturing the vision-based components of the sketches, similar to what a human evaluator may use, we hope to enable software that performs similarly to expert level.

Our neural network achieved generalization and validation accuracies of 75.0% and 66.7%, respectively, with a set of curve drawings, and a generalization and validation accuracy of 72.3% with the corner dataset. By comparison, we also asked two human evaluators to label age groups based on a portion of the dataset, and they achieved 77.5% and 66.7% for the curve and corner drawings, respectively. In general, the human judges struggled more with corner drawings, and these findings provide support for future work integrating more sketch recognition features, such as pressure and timing information, in order to make better assessments that are fast, easy, and accessible for parents and children.

AU - Aniket Patel
AU - Seth Polsley
AU - Tracy Anne Hammond
CY - Prairie View, Texas
DA - 2022/03/16
PB - ASEE Conferences
TI - Using Neural Networks to Distinguish Children’s Age with Visual Features of Sketches
UR - https://peer.asee.org/39220
DO - 10.18260/1-2--39220
ER -