1. Posture monitoring

The alignment and dynamics of human posture play a fundamental role in maintaining musculoskeletal health. Prolonged suboptimal postures and movement patterns increase the risk of musculoskeletal disorders (MSD). MSDs develop from repeated mechanical stresses – including strain, compression, and rotational forces – on muscles and skeletal structures. Such issues are particularly prevalent in workplace environments, where employees maintain static positions for prolonged durations (Singh and Singh, Reference Singh and Singh2019). Environmental factors, such as workplace ergonomics and furniture configuration, can significantly contribute to postural deterioration. Beyond the workplace context, specific populations like wheelchair users face unique challenges, with research indicating a high prevalence of spinal complications (Liampas et al., Reference Liampas, Neophytou, Sokratous, Varrassi, Ioannou, Hadjigeorgiou and Zis2021) due to sustained seated positions and mobility device design constraints.

The increasing digitalization of work environments and their impact on postural health forms the cornerstone of this investigation, with particular attention to seated positions commonly adopted during professional activities. The modern workplace has witnessed a surge in office syndrome manifestations, primarily attributed to extended computer-based tasks. This trend is further exacerbated by the ubiquitous use of mobile devices, leading to a phenomenon known as “tech neck” – a condition characterized by chronic neck strain from prolonged downward head positioning. Given these emerging health challenges, the systematic monitoring and evaluation of spinal alignment, particularly in the cervical and thoracic regions, has become essential for preventing musculoskeletal complications associated with contemporary lifestyle patterns. As illustrated in Figure 1, a notable study was conducted by Yu et al. (Reference Yu, James, Edwards and Snodgrass2018) to evaluate postural variations across different computing device configurations. Their research revealed that portable devices, specifically notebooks and tablets, induced more significant postural deviations – including increased neck flexion, bilateral shoulder elevation, and upper trunk flexion – compared to traditional desktop setups.

Figure 1.

Posture monitoring experiment (reproduced from Yu et al., Reference Yu, James, Edwards and Snodgrass2018).

Healthcare practitioners in ergonomics, movement science, and rehabilitation rely heavily on quantifiable musculoskeletal metrics, particularly those derived from joint range-of-motion measurements. Traditional assessment methods include manual goniometry (Hancock et al., Reference Hancock, Hepworth and Wembridge2018), which requires direct physical contact for angle measurement, and image-based analysis software such as Kinovea (Fernández-González et al., Reference Fernández-González, Koutsou, Cuesta-Gómez, Carratalá-Tejada, Miangolarra-Page and Molina-Rueda2020). While these approaches excel in point-in-time evaluations, they prove inadequate for continuous monitoring in workplace settings. Software tracking points often drift from their correct positions over time. As a result, evaluators are required to manually adjust tracking points throughout the session, making long-term posture monitoring cumbersome and time-consuming (Salisu et al., Reference Salisu, Ruhaiyem, Eisa, Nasser, Saeed and Younis2023).

Modern research in continuous posture monitoring predominantly employs sophisticated 3D motion capture technology (Yu et al., Reference Yu, James, Edwards and Snodgrass2018; Lee et al., Reference Lee, James, Edwards and Snodgrass2021; Lee et al., Reference Lee, James, Edwards and Snodgrass2022). This methodology typically involves strategically positioning infrared-reflective markers on anatomical landmarks, with multiple cameras capturing synchronized imagery to reconstruct three-dimensional positional changes over time. Despite the precision and comprehensive spatial data offered by 3D motion capture systems, several practical limitations restrict their widespread adoption. These systems demand substantial financial investment, require meticulous calibration procedures, necessitate controlled laboratory environments, and depend on specially trained personnel for operation and data interpretation (Obukhov et al., Reference Obukhov, Volkov, Pchelintsev, Nazarova, Teselkin, Surkova and Fedorchuk2023).

In light of these challenges with existing equipment, the researchers aim to develop a posture tracking system that simplifies and enhances usability. The proposed system should efficiently track posture, consistently monitoring desired points without losing tracking from markers. Since a single camera’s viewpoint is limited, integrating sensors helps compensate for body parts that are not easily visible, ensuring more comprehensive tracking. This improvement will eliminate the need for manual adjustments by evaluators during sessions, making long-term posture monitoring more convenient. Additionally, the system should be affordable, easy to use, and uncomplicated, allowing the employees to utilize it independently while facilitating data transfer to specialists for further analysis. This research introduces Posture Lab, a prototype system that enables real-time posture monitoring through accessible technologies. The implementation combines computer vision techniques utilizing ArUco markers, fiducial markers traditionally employed in robotic navigation (García-Ruiz et al., Reference García-Ruiz, Romero-Ramirez, Muñoz-Salinas, Marín-Jiménez and Medina-Carnicer2023), with wearable accelerometer sensors (Greene et al., Reference Greene, Doheny, McManus and Caulfield2022; Ali et al., Reference Ali, Arjunan, Peter, Perju-Dumbrava, Ding, Eller, Raghav, Kempster, Motin, Radcliffe and Kumar2023).

The Posture Lab focuses on two key measurements: Normal head posture (NHP) versus forward head posture (FHP) and normal thoracic spine versus kyphosis measurements. The craniovertebral angle (CA) and shoulder angle (SA) are used for measuring NHP and FHP, while the kyphosis angle (KA) assesses normal thoracic spine alignment and thoracic kyphosis. The Posture Lab employs computer vision technology with ArUco marker detection via a webcam to analyze head positions. Additionally, wearable accelerometer sensors measure kyphosis by checking the angle of inclination. The framework includes a web-based user interface for registration and specialized desktop applications for various measurement protocols. All components of the system communicate through a RESTful API, with data stored centrally for comprehensive reporting. Employees with basic musculoskeletal assessment knowledge can operate the Posture Lab. While acknowledging potential trade-offs in measurement precision, the Posture Lab prioritizes accessibility and practical implementation as a primary healthcare screening tool aimed at facilitating early detection of potential musculoskeletal complications.

Furthermore, in this study, we compared Posture Lab with Kinovea, as both systems employ single-camera 2D analysis. The aim was to evaluate the feasibility of using Posture Lab – a single camera system with markers and accelerometer sensors – to monitor human sitting postures. While Kinovea’s reliance on manual intervention limits its practicality for extended-duration assessments, it remains a convenient, valid and reliable field-testing device due to its single-camera setup and high accuracy compared to 3D motion analysis (Shishov et al., Reference Shishov, Elabd, Komisar, Chong and Robinovitch2021; Puig-Diví et al., Reference Puig-Diví, Escalona-Marfil, Padullés-Riu, Busquets, Padullés-Chando and Marcos-Ruiz2019). If the data from Posture Lab align closely with Kinovea’s results, it would suggest that Posture Lab could serve as a viable alternative for postures analysis, particularly in scenarios where single camera 2D systems like Kinovea are traditionally used.

2. Posture measurement techniques

The assessment of posture-related musculoskeletal disorders involves examining multiple anatomical deviations, particularly in the cervical, shoulder, and spinal regions. These physical alterations are routinely evaluated in clinical settings by physical therapy professionals. FHP and kyphosis angle have emerged as particularly valuable indicators, especially for office workers who frequently lean forward toward computer screens and wheelchair users who maintain prolonged seated positions. These metrics are primarily selected due to their quantifiable nature using conventional assessment techniques. However, existing measurement approaches present challenges for continuous monitoring needed to evaluate behavioral patterns or environmental influences in both workplace settings and daily wheelchair use. Based on these considerations, our prototype system specifically targets these two key postural parameters, which are critical indicators for both population groups.

Under optimal conditions, the head maintains vertical alignment with the spine, characterized by ear–shoulder vertical correspondence. FHP (Chu, Reference Chu2022) describes a postural deviation where the cervical spine alignment is compromised, resulting in anterior head displacement. This condition frequently develops from sustained forward head tilt during computer or mobile device usage, potentially leading to muscular and skeletal stress in the cervical and upper thoracic regions. Maintaining FHP for extended periods can cause pain, stiffness, and headaches. Clinical assessment of FHP traditionally relies on observing the ear–shoulder–hip vertical alignment. Optimal posture is characterized by these anatomical landmarks forming a vertical reference line, with deviations from this alignment suggesting potential FHP development. According to contemporary research works, the CA typically measures less than 48° (Kim et al., Reference Kim, Kim and Son2018), signaling postural dysfunction and SA often exceeds 54° (Nam et al., Reference Nam, Son, Kwon and Lee2013), reflecting compensatory muscular adaptations. This postural syndrome can lead to increased cervical spine strain, muscular imbalances, and potential long-term musculoskeletal complications.

Our assessment methodology quantifies FHP severity through the forward head angle measurement, defined by the angular relationship among three anatomical landmarks: the tragus (ear), C7 vertebral spinous process, and acromion process of the scapula, while conventional assessment methods employ goniometry or image analysis software like Kinovea as shown in Figure 2. Although Aliaa (Reference Aliaa2016) concluded in their study that Kinovea demonstrated Kinovea’s practicality for field studies and demonstrated validated Kinovea’s utility for large-scale postural studies. Moreover, Sharifnezhad et al. (Reference Sharifnezhad, Raissi, Forogh, Soleymanzadeh, Mohammadpour, Daliran and Bagherzadeh Cham2021) concluded in their study that Kinovea demonstrates excellent interrater and intrarater repeatability for measuring kyphosis and lumbar lordosis. However, the automatic movement tracking of the Kinovea program often encounters issues with tracking the desired anatomical landmarks. It is frequently observed that the tracking deviates from the correct points, necessitating manual tracking to correct these errors. Additionally, when tracking postures over extended periods, such as analyzing sitting posture during work, there tends to be a need for manual data analysis. This can be time-consuming due to the large amount of data (Salisu et al., Reference Salisu, Ruhaiyem, Eisa, Nasser, Saeed and Younis2023). However, while Kinovea’s reliance on manual intervention limits its practicality, especially for extended-duration assessments, it remains a convenient, valid, and reliable field-testing device due to its single-camera setup and high accuracy compared to 3D motion analysis, which is considered the gold standard (Shishov et al., Reference Shishov, Elabd, Komisar, Chong and Robinovitch2021; Puig-Diví et al., Reference Puig-Diví, Escalona-Marfil, Padullés-Riu, Busquets, Padullés-Chando and Marcos-Ruiz2019).

Figure 2.

Measurement of forward head angle using Kinovea.

To address the challenge of identifying subcutaneous landmarks, our approach employs a practical solution where individuals with basic training in anatomical landmarks can physically locate these points through palpation, then mark them with fiducial tags. We selected ArUco markers as the visual reference system due to their superior detection reliability compared to AprilTag, another popular fiducial tag (Zakiev et al., Reference Zakiev, Tsoy, Shabalina, Magid and Saha2020). These markers serve as clear visual anchors that can be consistently tracked using computer vision techniques, bridging the gap between manual anatomical identification and automated measurement. The markers, once placed on identified landmarks, enable continuous tracking of postural changes through standard webcam technology. Based on consultation with physical therapists, we selected a marker size of 1 square centimeter – large enough for webcam detection while minimizing coverage of anatomical landmarks. Through experimental testing of various ArUco dictionaries – predefined sets of square markers with different bit patterns – printed as the grid on standard A4 papers, we determined as shown in Figure 3 that the 4 × 4–50 dictionary provided optimal detection at typical experimental distances (1–2 m).

Figure 3.

Experiments for evaluating ArUco markers.

Detecting ArUco markers from images captured by webcams is achieved through Python code that utilizes the OpenCV library. The ArUco API provides the list of detected ID and (x, y) coordinates of all four corners for each marker. These corner coordinates can be used to calculate the center of the marker, as shown in (1) and (2). The system then processes these marker positions to compute the forward head angle, providing quantitative assessment of FHP severity.

(1) $$ {x}_{center}=\frac{1}{4}\sum \limits_{i=1}^4{x}_i $$

(2) $$ {y}_{center}=\frac{1}{4}\sum \limits_{i=1}^4{y}_i $$

Let ( $ {x}_{C7} $ , $ {y}_{C7} $ ) and ( $ {x}_{tragus} $ , $ {y}_{tragus} $ ) represent the 2D coordinates of the C7 spinous process and ear tragus markers, respectively. The horizontal vector $ {v}_{horz} $ is defined as (1,0), aligned with the image’s x-axis. The vector from C7 to the tragus is $ {\mathbf{v}}_{tragus-C7} $ = ( $ {x}_{tragus}-{x}_{C7} $ , $ {y}_{tragus}-{y}_{C7} $ ). Then, the CA angle is computed as:

(3) $$ {\theta}_{CA}={\cos}^{-1}\left(\frac{{\mathbf{v}}_{horz}\cdot {\mathbf{v}}_{tragus-C7}}{\parallel {\mathbf{v}}_{horz}\parallel \cdot \parallel {\mathbf{v}}_{tragus-C7}\parallel}\right) $$

The SA is derived similarly using shoulder marker coordinates. Let ( $ {x}_{shoulder} $ , $ {y}_{shoulder} $ ) denote the 2D coordinates of the acromion markers. The SA is computed as:

(4) $$ {\theta}_{SA}={\cos}^{-1}\left(\frac{{\mathbf{v}}_{\mathrm{horz}}\cdot {\mathbf{v}}_{\mathrm{shoulder}-\mathrm{C}7}}{\parallel {\mathbf{v}}_{\mathrm{horizontal}}\parallel \cdot \parallel {\mathbf{v}}_{\mathrm{shoulder}-\mathrm{C}7}\parallel}\right) $$

Kyphosis, characterized by an excessive outward curvature of the upper thoracic spine (between the neck and ribs), is another posture-related disorder often caused by prolonged poor posture, such as slouching or hunching over for extended periods. If left unattended, kyphosis can lead to deformities in the upper back, resulting in a curved or “humpback” appearance as shown in Figure 4.

Figure 4.

Musculoskeletal disorder (MSD) associated with kyphosis.

The curvature of the upper back (thoracic spine) can be assessed using various indicators, such as the kyphosis angle and alpha/beta angles, with the kyphosis angle being the most commonly used technique. Traditionally, the evaluation of kyphosis has often relied on X-ray images of the spine. However, this approach is not suitable for continuous monitoring of back curvature during routine activities or in workplace environments. Therefore, there is a need for alternative methods that can accurately assess kyphosis without the limitations of X-ray imaging.

The Posture Lab chose to measure the kyphosis angles (KAs) formed by the prominences at T1 and T2 (alpha angle) and at T12 and L1 (beta angle), and then add them together to obtain the desired kyphosis angle as shown in Figure 5. Although the calculation of these two angles does not directly reflect the curvature of the back, a study (Lewis and Valentine, Reference Lewis and Valentine2010) found that the alpha, beta angles are correlated to the kyphosis angle. Our survey revealed that accelerometer sensors have been employed as digital inclinometers for measuring the alpha/beta angle, which is then presented as the kyphosis angle in both smartphone applications (Huang et al., Reference Huang, Zhao, Wang, Zhang, Zhao, Xiong, Zhong and Luo2022) and products such as EasyAngle. Therefore, the Posture Lab chose to use accelerometer sensors as wearable devices, allowing for continuous measurement and monitoring of the alpha and beta angles corresponding to the relevant bone landmarks. For kyphosis assessment in this study, the Posture Lab employs two wireless inclinometers which wirelessly connect to a computer via Bluetooth 5.0 protocol. The accelerometer sensors are a compact device featuring an MPU9250 as a 9-axis inertial measurement unit (IMU) with a built-in battery. The tilt angles relative to the Earth’s axis are calculated from the acceleration data along the X-axis, sampled at a rate of 1 Hz. The inclination accuracy is specified as 0.2°.

Figure 5.

Measuring alpha and beta angles with accelerometer sensors to analyze kyphosis (reproduced from Huang et al., Reference Huang, Zhao, Wang, Zhang, Zhao, Xiong, Zhong and Luo2022).

The approaches of ArUco markers and wearable devices enable the assessor to determine how the employees’ posture and behavior change over time while sitting or standing to perform assigned tasks. The continuous monitoring capability provided by wearable accelerometer sensors offers a significant advantage over traditional methods that rely on static measurements or snapshots.

3. System architecture

This research presents a prototype system – Posture Lab – for workplace musculoskeletal disorder (MSD) assessment, designed specifically as an accessible primary healthcare service within organizational settings. The system architecture prioritizes privacy protection, user-friendly operation, and cost-effective implementation while serving three key stakeholder groups:

• • Organizations: Enables proactive workplace health management through systematic posture monitoring, supporting occupational health initiatives without significant infrastructure investment.

• • Assessment Personnel: Accommodates staff with basic training in musculoskeletal assessment protocols, focusing on screening and data collection rather than clinical diagnosis, thereby reducing operational costs and training requirements.

• • Employees: Provides nonintrusive posture evaluation during regular work activities, ensuring minimal disruption to daily tasks while maintaining data privacy.

Given the need to handle multiple simultaneous assessments while protecting personal information, the system implements a strategic separation between confidential registration data and posture evaluation processes.

The flow of system interactions, illustrated in Figure 6, comprises two primary components designed for efficient healthcare service delivery. The registration component operates through a web-based platform, providing secure handling of personal information and assessment scheduling. The evaluation component runs locally on assessment station computers connected to accelerometer sensors, ensuring data privacy and reducing network security risks. These components communicate through a secure RESTful API, with all data centrally stored to facilitate comprehensive reporting while maintaining data protection standards.

Figure 6.

Flow of system interactions.

As shown in Figure 7, the registration system leverages Docker containerized architecture for cost-effective deployment and maintenance:

• • User interface container utilizing Python with Streamlit and FPDF libraries, ensuring intuitive operation for nontechnical healthcare staff.

• • API container employing FastAPI for secure, efficient data exchange.

• • MongoDB container providing scalable, privacy-compliant data storage.

Figure 7.

Local web-based software architecture.

The database structure incorporates three distinct collections, segregating personal information, assessment requests, and evaluation results to enhance data security. The API implementation features two dedicated channels: a GET channel for assessment station job assignments and a POST channel for results submission. System deployment utilizes Docker Compose for streamlined installation and operation, minimizing IT support requirements and maintenance costs.

The assessment stations feature dedicated desktop applications developed using Python with PySide6, each tailored to specific postural measurements. The FHP assessment station utilizes webcam technology to track ArUco markers for forward head angle calculations, while the kyphosis station employs wireless accelerometers to measure gravitational inclination for alpha/beta angle determinations.

Figure 8 outlines the user journey through the assessment process. Employees begin by either completing initial registration or accessing their existing profile through a previously assigned identifier. After selecting their desired assessment type, the evaluation station retrieves their information and displays it for the attending staff member. The assessment process emphasizes clear communication, with staff members providing detailed explanations of procedures before beginning measurements. During the evaluation period, participants perform specified activities while the system continuously monitors their postural metrics, with real-time trending displayed for the evaluator’s reference.

Figure 8.

The assessment process.

Upon completion of the assessment, assessor conducts a preliminary analysis of the collected posture data and provide initial recommendations based on observed patterns. This information is securely transmitted to the central system through the REST API. The process concludes with the generation of a comprehensive PDF report at the registration area, detailing the assessment findings.

This document serves dual purposes: providing immediate feedback to the employees while also serving as a standardized reference for healthcare professionals, such as physical therapists, movement scientists or orthopedic specialists, who may conduct more detailed clinical evaluations including advanced imaging studies.

The Web-based User Interface for Registration Stage, illustrated in Figure 9, demonstrates the system’s user-centric approach. The web-based platform guides users through a streamlined process, differentiating between first-time and returning employees. New users complete a profile including essential health metrics such as weight and height, receiving a unique identifier for future visits. Returning users can efficiently access and update their existing information before proceeding to schedule their assessment.

Figure 9.

Web-based UI for registration stage.

The prototype system of Posture Lab features FHP and kyphosis assessment stations as shown in Figure 10(a) The FHP assessment station, which is connected to a Logitech C920 webcam with a resolution of 2048 × 1080 pixels. The webcam is positioned in portrait orientation and placed approximately 1 m away from the person under assessment and the kyphosis assessment station using two WitMotion WT9011DCL-BT50 devices as wireless inclinometers. Two tasks (reading documents on iPad and notebook computer) were given as the computer software at the FHP and kyphosis assessment stations import the video feed and detects ArUco markers and accelerometer sensors. The positions of the ArUco markers are at numbered 1 (ear position: tragus of the ear), numbered 2 (lower neck position: spinous process of C7 vertebra), and numbered 3 (shoulder position: acromion process of the scapula). For the kyphosis assessment station, the positions of the two wireless inclinometers are at numbered 4 (upper back position: T1 and T2 vertebrae) and numbered 5 (mid-back position: T12 and L1 vertebrae) as shown in Figure 10(b). Using the ArUco markers and accelerometer sensors, the software calculates and continuously displays the forward head angle and kyphosis on the screen every second and the evaluation time can be custom-selected. The user interface of FHP assessment is displayed in Figure 11(a), which shows FHP data on a screen. On the right side of the screen, there is a graph with the x-axis representing CA (red line) and SA (blue line) in degrees and the y-axis representing time in seconds. The left side of the screen features an image where the system creates a hypothetical line (blue lines) from the markers at the ear position to the lower neck position and connects to the shoulder position. Additionally, it generates another hypothetical line (green lines), which is a horizontal plane cutting between the marker positions at the lower neck and shoulder, forming angles CA and SA. Moreover, the user interface of kyphosis assessment is displayed in Figure 11(b), which shows kyphosis data on a screen. On the right side of the screen, there is a graph with the x-axis representing alpha (red line), beta (blue line), and kyphosis (green line) angles in degrees and the y-axis representing time in seconds. The left side of the screen features an image; however, the system does not create any hypothetical line. The Posture Lab outputs an assessment report with graphs and images and raw data in Excel files for CA, SA, and KA values, allowing experts to analyze FHP and kyphosis further.

Figure 10.

(a) FHP and kyphosis assessment station, (b) position of the markers and accelerometer sensors: (1) ear, (2) lower neck, (3) shoulder, (4) upper back, (5) mid-back positions.

Figure 11.

User interface of (a) FHP and (b) kyphosis assessment.

The example of an assessment report, as shown in Figure 12, is developed using the FPDF library based on data and assessment status. The collected data are aggregated on a per-minute basis and presented in the form of a box plot, displaying the trend of indicator changes over time. Additionally, a histogram is included to illustrate the distribution of indicator values. Compared to traditional snapshot-based posture indicator measurements, the inclusion of trend and distribution information enables healthcare professionals to assess the progression of MSD and posture status in near-real-life environments. This comprehensive information allows for a more informed evaluation, leading to the development of effective intervention for behavior modification or appropriate treatment recommendations.

Figure 12.

Example of assessment report.