Can We Prevent "Text Neck" Using Only a Smartphone? Real-Time Neck Angle Estimation and a Serious Game as a Case Study
>100 Views
December 02, 25
スライド概要
Prolonged smartphone use often results in forward head posture, commonly known as "text neck," which places excessive strain on the cervical spine and may lead to various health problems. However, users typically find it difficult to monitor and correct their posture without external support. To address this challenge, we propose a method for real-time neck angle estimation using only a smartphone’s front-facing camera and built-in sensors. Our approach extracts posture-related features such as the normalized distance from the nose to the neck base, interocular distance, smartphone tilt, and facial orientation. Regression analysis using Random Forest and Extra Trees models indicates that neck angle can be estimated with moderate accuracy (for example, 𝑅2 ≈ 0.6 and MAE ≈ 8.5◦), although performance at extreme angles remains limited. To evaluate the practical utility of this system, we developed a serious game titled Look Up and Tap! that provides real-time visual feedback, including screen contrast adjustment and button highlighting, based on the estimated neck angle. A user study with 10 participants showed that this feedback significantly improved users’ awareness of their posture and encouraged more upright neck positions. These findings suggest that neck angle estimation using only a smartphone is feasible and can help promote posture awareness in everyday settings. The serious game serves as one example of how this approach can be applied in real-world scenarios.
明治大学 総合数理学部 先端メディアサイエンス学科 中村聡史研究室
（ダウンロード不可）
関連スライド
各ページのテキスト
Can We Prevent “Text Neck” Using Only a Smartphone? Real-Time Neck Angle Estimation and a Serious Game as a Case Study Kento Watanabe, Satoshi Nakamura Meiji University, Japan
Background Q. What is your usual posture when using a smartphone? 1
Background 2
Background: “Text Neck” • Many people use smartphones while maintaining a Forward Head Posture (FHP). • When sustained for long periods, this posture can lead to “text neck”, or a straightened cervical spine. Neck angle 0° 15° 30° 45° * 60° Neck load 5kg 12kg 18kg 22kg 28kg * Hansraj, K. K.. Assessment of Stresses in the Cervical Spine Caused by Posture and Position of the Head. Surgical Technology International, 2014, vol. 25, no. 25, pp. 277-279. 3
Background: System Necessity Difficult to maintain continuous self-awareness A system that estimates smartphone posture in realtime and provides feedback can help support better posture habits. 4
Related Work Measuring neck angle using accelerometers attached to the neck or the back of the neck e.g., earphones [Liao+ 2016], necklaces [Chung+ 2019], etc. High introduction cost (sensor purchase & attachment) Estimating neck angle only from the smartphone [Lee+ 2013, Lawanont+ 2015, 2018] Accurately estimated in limited condition Liao, D. Y. et al. “Design of a Secure, Biofeedback, Head-and-Neck Posture Correction System.” Proc. IEEE CHASE, 2016, pp. 119–124. Chung, H. Y. et al. “Design and Implementation of a Novel System for Correcting Posture Through the Use of a Wearable Necklac e Sensor.” JMIR Mhealth Uhealth, vol. 7, no. 5, 2019, e12293. Lee, H. et al. “A New Posture Monitoring System for Preventing Physical Illness of Smartphone Users.” Proc. IEEE CCNC, 2013, pp. 713–716. Lawanont, W. et al. “Smartphone Posture Monitoring System to Prevent Unhealthy Neck Postures.” Proc. JCSSE, 2015, pp. 331–336. 5
Purpose Estimate neck angle only with a smartphone and provide real-time feedback to encourage posture correction 6
Neck Angle Estimation Method Features: 1. Phone orientation 2. Face angle 3. Landmark alignment 7
Dataset Construction • Continuously collected data for each parameters • 84,374 samples from 5 participants (4 male, 1 female) Neck angle Face angle Phone angle Phone-face distance 8
Estimation Result • Trained a regression model and evaluated its accuracy • 𝑹𝟐 : 0.61, 𝑴𝑨𝑬: 8.68, 𝑹𝑴𝑺𝑬: 11.64 9
Estimation Result • Trained a regression model and evaluated its accuracy • 𝑹𝟐 : 0.61, 𝑴𝑨𝑬: 8.68, 𝑹𝑴𝑺𝑬: 11.64 10
Estimation Result • Trained a regression model and evaluated its accuracy • R²: 0.61, MAE: 8.68, RMSE: 11.64 11
Case Study: Serious Game • Poor posture → harder Good posture → easier • Highlight next target when “good posture” → Encourages maintaining correct posture 12
User Study With vs. without posture-based control (n=10) Subjective evaluation after each condition • Significant improvement in posture awareness (p < .01) • More time above “good posture” threshold (p < .01) Self-posture awareness Proportion “good posture” 13
Package and Applications * Published JS library and held a 7hours small hackathon Watching 360º video with neck angle Fishing game using neck angle * https://github.com/nkmr-lab/neck-pose-estimator 14
Summary • Proposed neck angle estimation using only a smartphone • Built a continuous dataset with multiple parameters • Trained Regression model → moderate accuracy • Applied to serious game for posture correction → posture awareness improved 15