Advances in Markerless Motion Capture Systems: A Review of OpenCap and Its Applications

Overview and Capabilities

OpenCap marks a major advancement in motion capture technology, employing sophisticated algorithms and multiple camera angles to monitor and analyze human movement without the need for the cumbersome physical markers used in traditional systems. It uses pose estimation algorithms to detect body landmarks from videos recorded with two or more smartphones and applies deep learning and biomechanical models to calculate three-dimensional kinematics and dynamics [4]. The system’s ability to predict dynamic measures such as muscle activations, joint loads, and joint moments enables its use in screening for disease risk, evaluating intervention efficacy, and informing rehabilitation decisions [4,5]. OpenCap’s accessibility and ease of use position it as a promising tool for large-scale human movement studies outside of traditional laboratory environments [4]. By embracing an open-source framework, OpenCap invites continuous innovation and customization, making it highly adaptable to the diverse and evolving needs of biomechanical research and clinical practice [4]. Additionally, OpenCap’s use of smartphones offers a more flexible, low-cost alternative, making it accessible for studies outside of traditional labs, which distinguishes it from more complex setups requiring extensive equipment.

Validation and Accuracy

Accuracy is a pivotal criterion in evaluating OpenCap, especially in comparison to the well-established precision of traditional marker-based systems. OpenCap has demonstrated a rotational mean absolute error (MAE) of 4.5 degrees and a translational MAE of 12.3 mm across various movements, including walking, squatting, sit-to-stand, and drop jump, compared to a traditional marker-based system [4]. Further research extended this validation to assess concurrent accuracy and inter-trial variability in pathological gait patterns, revealing an average root mean square error (RMSE) of 5.8 degrees and greater variation in lower extremity kinematics across walking trials compared to a marker-based motion capture system [6,7]. Additionally, OpenCap has shown a MAE of 3.85 degrees and a RMSE of 4.34 degrees across lower extremity joint angles during common return-to-sport tasks involving hopping and landing techniques [5]. However, traditional systems, which typically achieve RMSEs of less than 2.0 mm for moving markers and 1.0 mm for stationary markers, are recognized for their superior spatial accuracy [8].

Error values from bone-pin and radiographic systems support these findings [9-14]. Discrepancies between marker-based systems and direct measurements have been reported, with RMS differences for marker positions ranging from 1.06 mm to 8.31 mm depending on the task and plane of motion [13]. Specifically, sagittal plane errors are smaller, with 2.45 degrees for walking and 2.67 degrees for running, while transverse plane errors are larger, reaching up to 4.70 degrees for running. Surface movement errors of up to 10.5 mm in knee joint center locations and leg rotational errors of up to 8 degrees for joint kinematics have been observed [11]. These limitations affect the validation of markerless systems like OpenCap, which are typically benchmarked against traditional systems.

Challenges in traditional systems include marker placement on anatomical landmarks and variability due to day-to-day differences and inter-tester inconsistencies [9,15]. In complex movements, slight discrepancies in marker placement significantly affect data accuracy [10]. Markerless systems like Open-Cap may reduce these issues by eliminating the need for physical markers, minimizing errors introduced by human variability.

Challenges and Future Directions

While OpenCap eliminates the need for physical markers, offering a less intrusive method of motion capture, it faces challenges such as increased inter-trial variability. This variability necessitates improved data processing to achieve levels of accuracy comparable to traditional systems [6,7]. This has important implications for post-processing, where averaging waveforms is recommended to mitigate the effects of this variability, as opposed to relying on single trials that may lead to misleading conclusions. Additionally, while calibrationless monocular vision systems have been investigated as simpler alternatives for motion capture, they generally result in larger errors compared to multi-camera systems like OpenCap, indicating they may not yet be suitable replacements in applications requiring high precision [16].

OpenCap’s preset video capture configuration, which utilizes a sampling frequency of 60 Hz and a resolution of 1280 x 720, is adequate for many biomechanical analyses, particularly those involving slower movements. However, it may not be as effective for capturing fast movements where higher frame rates would provide better accuracy. Future updates aimed at increasing the frame rate could enhance OpenCap’s performance in more dynamic contexts.

These findings suggest that while OpenCap provides reasonably accurate kinematic data across various movements, its precision is somewhat lower than that of traditional marker-based systems, particularly in dynamic activities. This variability in accuracy should be carefully considered when using OpenCap in settings where precise kinematic measurements are critical. Future research should focus on refining the algorithms used for pose estimation and enhancing data processing techniques specifically within markerless systems. Hybrid systems that combine markerless and marker-based technologies might offer potential advantages by leveraging the strengths of both approaches, though they could also introduce complexities, such as longer setup times and calibration challenges. For instance, hybrid systems using a Time of Flight camera, high-speed vision sensors, and inertial measurement units have been successfully applied to skiing and snowboarding, providing highly accurate real-time motion capture in dynamic outdoor environments [17,18]. Hybrid systems such as these offer advantages in complex, dynamic environments where traditional marker-based systems face limitations, including marker occlusion or environmental interference. Recent research demonstrated the effective use of hybrid coregistration of 2D video and 3D marker data in tasks where placing markers is impractical, such as juggling, highlighting the adaptability of these systems in different motion capture applications [19]. However, the increased complexity in setup, calibration, and data synchronization between systems poses challenges that must be managed. For example, while the accuracy of joint kinematics in these hybrid setups is high, aligning data from both marker-based and markerless sources requires meticulous calibration, which can offset the usability benefits gained from the markerless components. Advancing markerless technology alone may provide a more straightforward path to reducing errors and improving the reliability of data from these systems. Specifically, research could focus on developing more robust algorithms for dynamic activities or increasing sampling frequencies to capture rapid movements with greater precision.

The trajectory of OpenCap and similar markerless systems is promising, driven by continuous advancements in AI, computer vision, and sensor technology. The field must now address key challenges: reducing variability and determining the future role of hybrid systems in motion capture. These systems are expected to become increasingly accurate, user-friendly, and accessible [20]. Integrating real-time feedback and augmented reality could further enhance their utility in sport training and rehabilitation. However, addressing inter-trial variability through improved algorithms and data processing techniques will be crucial for broader adoption in clinical and sport settings [7]. Continued development of deep learning models and the expansion of databases to cover a broader range of populations and activities could have a significant impact on both research and clinical practices. OpenCap’s potential to democratize access to high-quality motion analysis could revolutionize biomechanical assessments [4].

Other Markerless Motion Capture Systems

While OpenCap represents a significant advancement in markerless motion capture, it is not the only system contributing to the development of motion analysis without the need for markers. Other systems, such as Theia3D (Theia Markerless, Canada), DeepLabCut (Mathis Lab, Switzerland), and Captury Live (The Captury, Germany), offer alternatives with different approaches to accuracy, usability, and real-world applications. Each system involves a trade-off between the number of cameras used and the complexity of the calibration process, which directly affects the balance between accuracy and ease of use.

Theia3D offers a highly accurate markerless motion capture system, with root mean square differences (RMSD) between joint centers of less than 2.5 cm for most joints when compared to traditional marker-based systems [20]. It excels in capturing spatiotemporal gait parameters and lower limb kinematics, particularly in the sagittal and frontal planes, with high agreement for joint angles such as flexion and extension [21]. However, challenges remain in measuring transverse plane movements, such as pelvic tilt and internal/external rotations, where RMSD can reach up to 13 degrees [20]. The system is particularly suited for clinical research due to its high accuracy and multi-camera setup, though the complexity of its setup and associated costs may present barriers for some users.

DeepLabCut (DLC) is an open-source, deep-learning-based markerless system designed for animal and human motion tracking [22]. Unlike OpenCap, which uses a multi-camera setup, DLC can perform effectively with a single camera after being trained on labeled data, providing flexibility for small-scale studies [23]. DLC’s application across diverse species and behaviors is made possible by its transfer learning approach, which allows for human-level accuracy with as few as 200 labeled images [22,23]. However, additional manual labeling may be required when the system encounters movements or subjects that deviate significantly from the training data, which may limit its generalizability. While DLC works well in simple or controlled environments, its performance tends to decline in dynamic or complex scenarios, particularly with occlusions or high-speed movements [22,23]. Its flexible camera setup and open-source platform make it accessible to a wide range of users, though the manual effort required for training can be time-consuming, especially for diverse or unstructured datasets.

Captury Live is a markerless motion capture system used in various fields, including biomechanics. It employs multiple cameras to deliver immediate feedback, making it especially useful for capturing real-time dynamic movements. However, its precision in recording joint angles and kinematics is generally lower than that of systems like OpenCap or Theia3D, particularly during complex movements. Studies have shown that while Captury Live reliably estimates variables such as jump length and separation distance ratios, its accuracy decreases for more detailed measurements, such as knee flexion or during dynamic movements like squats and jumps, especially when the knees are fully extended [24,25]. The system’s focus on real-time feedback and ease of use makes it well-suited for dynamic settings where immediate results are prioritized, though its lower accuracy compared to other systems may limit its use in detailed biomechanical analyses [24,25].

Applications in Biomechanics

In sport biomechanics, OpenCap has emerged as a valuable tool, particularly for analyzing athletic performance in environments that closely mimic real-world conditions, free from the constraints of traditional marker-based setups. By eliminating the need for cumbersome markers, OpenCap allows for the capture of kinematics and kinetics, providing data that is potentially more natural and accurate. In clinical settings, OpenCap offers a non-invasive method for assessing and monitoring patients. The system is increasingly being used in rehabilitation to track patient progress, assess movement deficiencies, and tailor interventions [4]. The ability to capture motion data without markers is especially beneficial for patients who may be uncomfortable with or physically unable to wear traditional markers [5].

However, OpenCap’s simplified modeling approach presents some limitations. It employs a 1-degree of freedom (DOF) model for the knee and a 2-DOF model for the ankle, which differs from the more complex 3-DOF models typically used in traditional marker-based systems. This simplification also means that a key clinical variable, knee valgus, is not captured by OpenCap’s default settings [4,5]. These limitations could restrict the system’s application in scenarios where detailed joint motion analysis or specific clinical assessments, such as knee valgus, are critical [26-28].

One of the most promising aspects of OpenCap is its potential for everyday use. Individuals can perform biomechanical assessments in their own homes using just a few cameras or even smartphones [4]. This accessibility opens new possibilities for fitness tracking, ergonomic assessments, and home rehabilitation, making advanced biomechanical analysis more widely available [3].

For large-scale studies and in-field applications, managing the increased inter-trial variability is crucial. Automated techniques for detecting and eliminating outliers before averaging data can improve the robustness of analyses, making OpenCap more viable for widespread use [7]. Additionally, OpenCap’s reliance on cloud computing for data processing enhances its accessibility, enabling users to collect, analyze, and visualize movement data without the need for specialized hardware or software [4].