Markerless motion capture vs. optical marker-based motion capture: As you conduct validation research in 2026, do you need a practical checklist to guide how to perform rehabilitation training and motion-capture analysis?
Release time:
2026-10-20 15:26
Source:
Markerless motion capture is set to experience an explosive wave of validation in 2026: almost every month, new studies compare it directly with optical marker-based systems. The conclusion isn’t that one will replace the other, but rather that a progressively clearer picture is emerging. Tiered Application List — Which scenarios can be used with confidence, which require a baseline, and which should simply be avoided. This article presents a checklist based on nine validation studies conducted in 2025–2026, followed by a framework for integrating the two types of data collection devices. Kinematic Biomechanical Analysis Workflow plan.
01 Classification of Motion Capture Equipment
TWO FAMILIES OF CAPTURE
Optical marker-based systems remain the gold standard: multiple cameras track reflective markers attached to the skin, and triangulation is used to reconstruct 3D coordinates, offering the highest accuracy and reliability; however, this approach is time-consuming due to the need for marker placement, and it is susceptible to occlusion and marker loss. [1] 。
Markerless systems skip the process of attaching markers: they use commodity RGB or depth cameras paired with 2D/3D human pose estimation networks (such as MediaPipe or Transformer architectures), categorized into monocular and multi-view types based on hardware, and rely on deep learning to directly infer joint positions from video. [1] Zero sticking points and zero clothing interference are its greatest appeal in rehabilitation settings.

Figure 1: Markerless multi-camera validation setup: a stereo camera array synchronized with an optical reference system (adapted from [1]).
However, “unlabeled” is not a uniform category. Even when analyzing the knee impact angle in lateral fall videos of elderly individuals, the performance of three pose estimation tools varied dramatically: the best-performing model had an average percentage error of 13.61±10.55%, while the worst reached 39.09±20.54%. [2] When selecting a model, the phrase “unmarked” in itself does not constitute any commitment to accuracy.
02 Use Case Analysis
A TIERED USABILITY LIST
Tier 1: Screening and longitudinal monitoring: Available. Synchronous dual-system treadmill testing in 30 patients with knee osteoarthritis showed intraclass correlation coefficients (ICCs) of 0.74–0.95 for cadence, speed, and stride length; the ICCs for normalized initial contact knee joint angles were approximately 0.67–0.74, but agreement in hip and ankle angles was generally lower. [3] — Temporal parameters and the knee sagittal plane angle can be determined using markerless methods, but distal joint motion cannot.

Figure 2: Bland–Altman consistency analysis of gait parameters between markerless and reference systems (adapted from reference [3])
Second layer: Controlled motion quality assessment: acceptable. For planar movements such as squats, the MediaPipe approach and the Kinovea 2D tool both yielded ICC values greater than 0.89 for all variables, with the highest agreement observed for the symmetry index of proportions (ICC = 0.956). [4] — Suitable for monitoring movement quality during rehabilitation training.
Third Layer: Macroscopic Displacement – Available Conditionally. At 25 Hz, a comparison between six consumer-grade action cameras and a Vicon motion-capture system yielded correlation coefficients of 0.916–0.994 for the displacement trajectories of 12 joints, with median normalized RMSE values ranging from 0.54% to 1.32%; however, this frame rate is insufficient to support high-precision impact analysis. [5] 。

Figure 3: Synchronous arrangement of the 25 Hz multi-camera markerless capture system and the Vicon reference system (adapted from [5])
Fourth Floor · Caution Area. The maximum flexion angle in upper limb movements is systematically underestimated, most notably at the shoulder joint. [6] ; For throwing—a type of full-body explosive movement—the RMSD range of joint angles between the unmarked and marked systems spans 7.17°–26.66° [7] ; Peak accelerations from climbing falls are underestimated, with the ankle and sacrum experiencing the most severe impacts [8] High-speed motion, end-effector joints, and out-of-plane movement are common limitations of current markerless methods.

Figure 4: Underestimation of peak acceleration by the markerless pipeline: maximum error at the ankle and sacrum (reproduced from reference [8])
The rule can be summarized in one sentence: Use proximal large joints, slow speed, and the sagittal plane; use distal small joints, high speed, and out-of-plane cautiously.
03 Sports Biomechanics Analysis Solution
MAKING BOTH ENDPOINTS WORK TOGETHER
The conclusion of the research on markerless gait verification using multiple cameras is that markerless gait analysis can serve as both an alternative to and a complement for conventional gait analysis. [9] — Provided you’re in the right position: optical markers handle reference‑grade, diagnostic‑level data capture, while markerless video is used for routine, high‑frequency screening; the two are periodically calibrated against each other.
SMART-DX EVO Distributed Intelligent Motion Capture System Reference-end setup: Multi-camera optical capture synchronized with force plates and electromyography on a common timebase, delivering diagnostic-grade kinematics. Notably, validating the accuracy of markerless studies inherently requires such an optical system as a reference—indeed, the reference system used in the aforementioned upper-limb validation study was an 8-camera SMART-DX system. [6] 。
The unmarked end is composed of BoB /Markerless access. According to the official definition, it is the interface between BoB products and SPLYZA Motion analysis software: SPLYZA Motion uses AI to estimate 3D motion from a single 2D video clip, enabling motion capture with a smartphone or webcam.
Enter BOB Subsequently, the motion data is mapped onto an embedded musculoskeletal model containing over 600 motor units, and inverse dynamics are used to compute joint torques—meaning that, in the end, a markerless video yields not skeletal animation but biomechanical quantities at the muscle level; historical match or therapy recordings can also be retrospectively reprocessed.
There is only one red line for the usage posture: unmarked results are used for Longitudinal comparison within the same device With screening stratification, conclusions involving absolute angles and impact loads are delegated to the optical reference end. Placing each end in its respective position transforms the checklist from one of constraints into one of efficiency.
The point of the applicable list is not to condemn unmarked motion capture, but to place it where it belongs.
References
[1] Reliability assessment of markerless technologies and stereophotogrammetry in human movement analysis. Frontiers in Sports and Active Living, 2026. DOI:10.3389/fspor.2025.1712332
[2] Validity of multiple human pose estimation tools for measuring knee impact angles in video-captured falls of older adults. PLoS ONE, 2026. DOI:10.1371/journal.pone.0335108
[3] Gait assessment using a 2D video-based pose estimation approach in knee osteoarthritis patients. Frontiers in Bioengineering and Biotechnology, 2026. DOI:10.3389/fbioe.2026.1869878
[4] A Direct Comparison Between Kinovea and a MediaPipe-based Markerless System for Deep Squat Analysis. Physical Therapy Rehabilitation Science, 2026. DOI:10.14474/ptrs.2026.15.1.46
[5] Evaluating a Multi-Camera Markerless System at 25 Hz Against a Vicon Reference for Basketball-Specific Movements. Sensors, 2026. DOI:10.3390/s26051689
[6] Accuracy and Reliability of Markerless Human Pose Estimation for Upper Limb Kinematics Across Ranges of Motion. Applied Sciences, 2026. DOI:10.3390/app16031202
[7] Comparison of Marker-Based and Markerless Motion Capture Systems in Throwing Movement Analysis. Biomechanics, 2025. DOI:10.3390/biomechanics5040100
[8] Validation of a Markerless Multi-Camera Pipeline Against IMUs for Peak Acceleration in Bouldering Falls. Sensors, 2026. DOI:10.3390/s26020662
[9] Verification of Markerless Gait Analysis: Multi-Camera Pose Estimation versus Marker-Based Systems. Medicina, 2026. DOI:10.3390/medicina62020418
Markerless motion capture,Tiered Application List,Motion capture,Biomechanical Analysis of Movement,Rehabilitation Training Assessment