JMIR Publications

ABSTRACT

Background:

Prolonged improper posture can lead to Forward Head Posture (FHP), contributing to disabilities such as headaches, impaired respiratory function, and fatigue. This issue is especially pertinent in driving scenarios, where individuals often maintain static postures for extended periods—a significant part of daily life for many. The development of a system capable of detecting FHP is crucial, as it would not only alert users to correct their posture but also serve the broader goal of contributing to public health by preventing the progression of chronic injuries associated with this condition.

Objective:

This paper presents a system that utilizes 3D human pose estimation to detect FHP from a single 2D image, specifically designed for use in driving scenarios—a common context in daily life. Accurate diagnosis of FHP typically requires dedicated devices, such as clinical postural assessments or specialized imaging equipment, but their use is impractical for continuous, real-time monitoring in various everyday settings. Thus, it is necessary to develop an accessible, efficient method for the regular assessment of posture that can be easily integrated into daily activities, providing real-time feedback and promoting corrective action.

Methods:

The system initially estimates 3D human anatomical keypoints from a given 2D image, utilizing the VideoPose algorithm. Subsequently, the system implicitly learns the underlying relationship between the estimated 3D keypoints and the correct posture (i.e., determining whether it is FHP or not) by employing recent deep neural network techniques.

Results:

The test accuracy was 0.7914 when inputs included all joints corresponding to the upper body keypoints. These experimental results affirm the feasibility of our approach, demonstrating that a single 2D image can offer significant insights into a user's posture. An ablation study further substantiated the contribution of individual joints to our model’s accuracy. Notably, omitting the right shoulder joint led to a high accuracy of 0.8207, underscoring its significant role in the model's predictive capabilities. Conversely, the exclusion of the right wrist joint resulted in the lowest accuracy at 0.6843, suggesting its critical contribution to accurate posture estimation. Additionally, the removal of the left hip joint showed a notable decrease in performance, affirming its utility in the model's assessment of FHP.

Conclusions:

Based on 2D image input using 3D human pose estimation joint inputs, it is possible to utilize deep learning to identify FHP and develop a posture correction system. We conclude the paper by addressing the limitations of our current system and proposing potential avenues for future work in this area.