Movement Outcomes Acquired via Markerless Motion Capture Systems Compared with Marker-Based Systems for Adult Patient Populations: A Scoping Review

Abstract

Mobile motion capture is a promising technology for assessing physical movement; markerless motion capture systems (MLSs) offer great potential in rehabilitation settings, given their accessibility compared to marker-based motion capture systems (MBSs). This review explores the current literature on rehabilitation, for direct comparison of movement-related outcomes captured by MLSs to MBSs and for application of MLSs in movement measurements. Following a scoping review methodology, nine databases were searched (May to August 2023). Eligible articles had to present at least one estimate of the mean difference between a measure of a physical movement assessed by MLS and by MBS. Sixteen studies met the selection criteria and were included. For comparison of MLSs with MBSs, measures of mean joint range of motion (ROM) displacement were found to be similar, while peak joint angle outcomes were significantly different. Upper body movement outcomes were found to be comparable, while lower body movement outcomes were very different. Overall, nearly two-thirds of measurements identified statistical differences between MLS and MBS outcomes. Regarding application, no studies assessed the technology with patient populations. Further MLS-specific research with consideration of patient populations (e.g., intentional error testing, testing in less-than-ideal settings) would be beneficial for utilization of motion capture in rehabilitation contexts.

1. Introduction

Across health contexts, access to rehabilitation services is often limited. Matching services to ever-increasing patient needs can be difficult, particularly with staffing challenges, waitlists, and barriers to in-person attendance [1]. During the COVID-19 pandemic, accessibility issues became even more pronounced, increasing the need to conduct remote and portable assessments via technology-supported virtual health services [2]. When providing rehabilitation services with the aid of technology, there are additional challenges to conducting comprehensive assessments [3]. One of those challenges is identifying clinical tools that can be used as portable and remote options and that are precise in assessing physical movements [4]. To meet this need, there has been a growing interest in the use of markerless motion capture systems (MLSs), given their efficiency in use of space and their potential ease of accessibility through handheld electronic devices [5]. MLSs capture three-dimensional movements without requiring markers [6]. Marker-based motion capture systems (MBSs) capture movement patterns from markers attached to a person’s body [4]. Although very precise, MBS systems require considerable space and expensive resources including multiple cameras and calibrated equipment, lengthy setup time, and the need for highly trained specialists to operate the equipment. Due to these practical limitations, MBSs are not easily accessed nor routinely used in rehabilitation settings, either virtually or in person, despite being the current gold standard for motion capture [5].

With the use of technology-supported rehabilitation for chronic diseases on the rise [7,8], there is great value in the use of portable technology-based tools to conduct remote assessments and to make access more efficient [9]. A recent systematic review explored the use of MLSs in clinical populations; that review made us wonder about the utility of using MLSs versus MBSs, and how physical measurements compare across these systems [10]. The rationale for this current review was to explore the research and outcomes that have directly compared MLSs and MBSs, in terms of the populations being tested, the measurable differences, and the ways in which the research itself was carried out, in order to better understand how physical movements are best captured and the accessibility of this technology. To our knowledge, no scoping reviews have been completed to date that have directly compared measures used in MLSs and MBSs, nor that have considered clinical application of MLSs during testing in general for adult patient populations [6,10]. Specifically, our aims were to understand the literature relating to (1) direct comparison of movement-related outcomes captured by MLSs and MBSs; and (2) recommended application of MLSs in measuring gross physical movements.

2. Methods

2.1. Study Design

This scoping review was guided by the most recent edition of the JBI Manual for Evidence Synthesis [11] and was conducted using Arksey and O’Malley’s proposed framework [12] as later refined by Levac et al. [13]. Arksey and O’Malley’s framework consists of five sequential steps for conducting scoping reviews: (1) identifying the research question, (2) identifying relevant studies, (3) selecting studies, (4) extracting and charting data, and (5) summarizing and reporting results [12]. In accordance with JBI recommendations, the PCC framework was used to identify the main concepts for the research question. PRISMA guidelines were followed throughout the search and article extraction.

Research Questions:

• How do movement-related outcomes captured by MLS compare with the same outcomes captured by MBS?
• What procedural characteristics of the use of MLS technologies are relevant to consider to inform future clinical research?

For this second question, we focused on the populations assessed, characteristics of the technology, and experimental setup and procedures. Our goal was to explore the measurement comparisons of how MLSs performed versus MBSs in adult patient populations, to better understand how to utilize these technologies from a physical rehabilitation perspective.

2.2. Authors

Our multidisciplinary study team included a graduate student researcher (M.P.), an expert physical therapist with specialization in rehabilitation technology (K.B.), an expert occupational therapist with specialization in rehabilitation technology (R.H.), a health sciences librarian (L.D.), two associate professors with backgrounds in clinical rehabilitation (occupational therapy and physiotherapy, respectively) and scoping reviews (N.D.D., S.B.), and a professor of medicine (P.T.).

2.3. Search Strategy

The original research question and search strategy were conceptualized and developed with the support of a health sciences librarian (L.D.), and through the collaboration of our study team (M.P., N.D.D., S.B., P.T.). Relevant search terms and multiple Medline searches were evaluated before determining the final strategy. A search of Medline, Embase, CINAHL Plus with Full Text (Ebscohost), Scopus, Web of Science (including Science Citation Index Expanded (SCI-EXPANDED), the Social Sciences Citation Index (SSCI), the Arts and Humanities Citation Index (A&HCI), the Emerging Sources Citation Index, and Engineering Village (Elsevier) databases for relevant literature published between January 2009 (inception of MLSs in research) and 10 August 2023 was conducted by a health sciences librarian (L.D.)). The search combined free text search terms for MLS with terms for other motion capture systems (i.e., (markerless or marker-less or Leap-motion-controller or Pose2Sim or Kinect or KinectV2 or DeepLabCut or Deep-lab-cut or Trazer) and (motion-capture or motion-tracking or motion-analys* or Optotrack or OptiTrack or Certus or Qualysis or Vicon or CodaMotion)). No language or study type filters were applied.

After conducting the initial searches, all results were imported into Covidence systematic review management software (a software platform used for collaborative literature reviews) [14].

2.4. Eligibility Criteria

Inclusion criteria consisted of the following requirements:

-

Population: Any non-athlete, human adult (aged 18 years and older);

-

Concept: Any voluntary movement-related measure assessed during a gross motor physical movement or task, including mobility of joints (functions of the range), muscle power functions (functions related to the force generated by the contraction of a muscle or muscle groups), and control of voluntary movement functions (functions associated with control over and coordination of voluntary movements) [15]. Kinematic variables, both joint-specific and whole-body, were eligible;

-

Context: Any study that included a measurable mean difference estimate of physical movement-related measure between at least one markerless motion capture system to at least one marker-based motion capture system;

-

Publication: Only full-text, peer-reviewed articles.

Exclusion criteria consisted of studies which did not meet the inclusion criteria and/or included the following considerations:

-

Mixed participant populations including individuals under 18 years of age;

-

Assessment of involuntary movements occurring during a voluntary task (such as evaluation of sway during a static posture/balance task, or compensatory response to perturbations);

-

Assessment of facial expressions;

-

Assessment of fine motor movements (such as finger movements and hand dexterity);

-

Correlation estimates or inference measures;

-

Assessment of athletes.

3. Study Selection

For the initial title and abstract screening of retrieved articles, each record was reviewed independently by two reviewers (M.P., N.D.D., S.B.). Reviewers met routinely to ensure cohesive screening and to resolve conflicts through consensus; the third reviewer was consulted to facilitate consensus as required. The next stage of the full-text review was completed independently by any two members of our review team (M.P., N.D., S.B., R.H., K.B.). A third reviewer supported consensus where discussion was required (M.P., N.D.D., S.B.). Excluded articles were tagged at this stage with their reason for exclusion.

4. Data Extraction

Two reviewers (M.P., N.D.D.) completed data extraction from Covidence to a Microsoft Excel-supported spreadsheet (Microsoft Excel, version 16.66.1). Extracted and charted data included the following categories: author, country, year, publication focus, participant demographics (sample size, age, biological sex assigned at birth, health status), technology details (type of MLS, type of MBS, number of MBS markers, number of MBS cameras, any additional equipment), procedures (type of instruction during assessment, inclusion of intentional error testing, camera positioning, number of assessors, etc.), functional movements assessed, measures and types of kinematic measure (i.e., joint-specific/whole-body), statistical techniques used, summary of findings, as well as any other observations.

5. Results

5.1. Search Results

The search across nine databases identified 5619 studies related to the research question; after removal of 2640 duplicates, 2979 articles remained for initial abstract and title screening. A total of 2730 articles were excluded at this stage, of which 249 full texts were subsequently reviewed. Of these, 16 articles met our selection criteria and were included in this review and extracted for use in our scoping review. Figure 1 depicts the flow diagram of the results of the search strategy.

Table 1. Summary of the study and population characteristics of the included studies.

Study Characteristics			Population Characteristics
First Author, Year [Ref]	Country	Journal Genre	Sample Size	Age in Years Mean (SD)	Sex (M:F)	Health Status
Cattaneo, 2023 [16]	Italy	Healthcare technology	5	-	-	Healthy
Grooten, 2018 [17]	Sweden	Clinical condition	30	42 (14)	16:14	8 (27%): lower back pain 22 (73%): healthy
Hu, 2021 [18]	China	Healthcare technology	30	22.8 (1.76)	8:22	Healthy
Lim, 2022 [19]	Republic of Korea	Healthcare technology	1	29	-	Healthy
Mauntel, 2021 [20]	United States	Rehabilitation, athletics, movement	20	20.5 (2.78)	10:10	Healthy
Mehrizi, 2018 [21]	United States	Healthcare technology	12	47.5 (11.3)	12:0	Healthy
Moro, 2022 [22]	Italy	Engineering, technology	16	27 (2)	10:6	Healthy
Neal, 2020 [23]	United Kingdom	Rehabilitation, athletics, movement	21	32.1 (12.9)	10:11	21 (100%): patellofemoral pain
Oh, 2018 [24]	United States	Healthcare technology	12	24.5 (6)	5:7	Healthy
Pashley, 2021 [25]	Australia	Healthcare technology	78	48.4 (16.5)	26:16 *	25 (32%): stroke 15 (19%): TBI, 2 (3%): prior neurosurgery 36 (46%): healthy
Pfister, 2014 [26]	United States	Healthcare technology	20	27.4 (10)	9:11	Healthy
Schmitz, 2015 [27]	United States	Rehabilitation, athletics, movement	15	24 (4)	8:7	Healthy
Skals, 2017 [28]	Denmark	Engineering, technology	10	23.5 (1.27)	10:0	Healthy
Tanaka, 2019 [29]	Japan	Healthcare technology	18	21 (0.6)	15:3	Healthy
Tanaka, 2019 [30]	Japan	Healthcare technology	60	20.9 (0.4)	41:18	Healthy
Yeung, 2021 [31]	China	Rehabilitation, athletics, movement	10	27.2 (4.7)	8:2	Healthy

[ref] = study reference; SD = standard deviation; Sex M:F = ratio of participants with male sex assigned at birth to participants with female sex assigned at birth; * = assigned sex at birth ratio for patients with acquired brain injury only.

summarizes the characteristics of the included studies and population samples. The 16 included studies [16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31] were conducted in ten different countries and were published between 2014 and 2023. Publication journal genres focused on healthcare technology for nine articles [16,18,19,20,21,24,25,26,28]; rehabilitation, athletics, and movement for four articles [20,23,27,31]; engineering technology for two articles [22,28]; and clinical conditions for one article [17]. In terms of study designs, all included studies employed a cross-sectional design. As for the type of statistical tests used to compare the measures from each system, the paired t-test was most frequently used [16,20,21,22,23,24,25,26,27,29,30,31], while the Wilcoxon ranked sum test was used in two studies [17,28], the Student’s t-test was used in one study [18].

5.2. Movement-Related Outcomes

To achieve our aim of understanding direct comparisons of movement-related outcomes captured by MLSs and MBSs,

Table 2. Summary of differences of movements and measures assessed by markerless and marker-based systems.

	Movement Plane	Physical Movement	Measure	Units	Smallest Difference between Systems	Greatest Difference between Systems	Statistically not Different (MLS = MBS) *	Statistically Different (MLS ≠ MBS) **	Systems Compared
Upper body	Frontal	Shoulder abduction/adduction	Joint ROM displacement	Degrees	0.30 [28]	2.60 [28]	2/3 [28]	1/3 [21]	Azure Kinect v. Optitrak [16] Kinect V2 v. Tracklab BTS Elite [17] 4×2D RGB Cameras v. Vicon MX [19] Kinect V2 v. Vicon Bonita [20] 2×2D RGB Cameras v. Motion Analysis Corp [21] Kinect V2 v. Optitrak [25] Kinect v. Oqus [28]
		Trunk side bending	Joint ROM displacement	Degrees	-	-	1/2 [17]	1/2 [17]
		Shoulder elevation/depression	Joint ROM displacement	Degrees	1.53 [25]	2.59 [25]	1/2 [25]	1/2 [25]
			Joint ROM displacement	Millimeters	-	-	-	2/2 [16]
			Peak joint angle	Degrees	2.2 [25]	2.68 [25]	2/2 [25]	-
		Trunk lateral flexion	Joint ROM displacement	Degrees	-	-	-	1/1 [16]
	Sagittal	Shoulder flexion/extension	Joint ROM displacement	Degrees	0.19 [28]	5.12 [25]	1/5 [21], 2/5 [25], 2/5 [28]	-
			Peak joint angle	Degrees	6.49 [25]	9.8 [25]	-	2/2 [25]
		Elbow flexion/extension	Joint ROM displacement	Degrees	1.6 [19]	22.37 [25]	1/4 [21], 1/4 [19]	2/4 [25]
			Peak joint angle	Degrees	4.8 [19]	29.01 [25]	-	2/3 [25], 1/3 [19]
		Trunk flexion/extension	Joint ROM displacement	Degrees	0.36 [20]	1.87 [21]	1/2 [20], 1/2 [21]	-
			Peak joint angle	Degrees	1.16 [20]	1.16 [20]	1/1 [20]	-
Lower body	Frontal	Hip abduction/adduction	Joint ROM displacement	Degrees	0.04 [20]	2.04 [20]	2/7 [20], 1/7 [21], 1/7 [22], 1/7 [28]	2/7 [20]	Kinect V2 v. Vicon MX [18,29,30] 4×2D RGB Cameras v. Vicon MX [19] Kinect V2 v. Vicon Bonita [20] 2×2D RGB Cameras v. Motion Analysis Corp [21] 3×2D RGB Cameras v. Optitrak [22] 2×RGB Cameras v. Codamotion [23] Kinect V2 v. BTS [24] Kinect v. Vicon MX [26] Kinect v. Motion Analysis Corp [27] Kinect v. Oqus [28] Azure Kinect; Kinect V2; Orbec Astra v. Vicon Bonita [31]
			Peak joint angle	Degrees	0.02 [20]	18.7 [31]	2/36 [20], 1/36 [23], 1/36 [27], 6/36 [31]	2/36 [20], 24/36 [31]
		Knee valgus/varus	Joint ROM displacement	Degrees	0.14 [20]	1.2 [20]	2/2 [20]	-
			Peak joint angle	Degrees	0.15 [20]	9.17 [18]	2/13 [20], 1/13 [27]	10/13 [18]
	Sagittal	Hip flexion/extension	Joint ROM displacement	Degrees	0.28 [20]	15.4 [19]	2/12 [20], 1/12 [21], 1/12 [22], 3/12 [24]	1/12 [24], 1/12 [28], 2/12 [30], 1/12 [19]
			Peak joint angle	Degrees	1.29 [20]	15.0 [31]	2/50 [20], 1/50 [24], 4/50 [26], 17/50 [31]	3/50 [24], 8/50 [26], 1/50 [27], 13/50 [31], 1/50 [19]
		Knee flexion/extension	Joint ROM displacement	Degrees	1.27 [20]	10.01 [28]	2/10 [20], 1/10 [21], 2/10 [24]	1/10 [22], 2/10 [24], 1/10 [28], 1/10 [19]
			Peak joint angle	Degrees	0.51 [18]	20.3 [31]	5/61 [18], 2/61 [20], 2/61 [24], 2/61 [26], 1/61 [27], 10/61 [31]	5/61 [18], 1/61 [23], 2/61 [24], 10/61 [26], 20/61 [31], 1/61 [19]
		Ankle plantarflexion/dorsiflexion	Joint ROM displacement	Degrees	1.17 [21]	37.88 [24]	1/9 [21]	1/9 [22], 4/9 [24], 1/9 [28], 2/9 [30]
			Peak joint angle	Degrees	11.56 [24]	21.17 [24]	1/24 [24], 1/24 [31]	3/24 [24], 19/24 [31]
		Pelvic tilt	Joint ROM displacement	Degrees	~2.0 [22]	~2.0 [22]	1/1 [22]	-
Whole body		Normal gait	Stride time	Seconds	0.01 [24]	0.04 [22]	2/20 [22], 1/20 [24]	5/20 [24], 12/20 [26]	3×2D RGB Cameras v. Optitrak [22] Kinect V2 v. BTS [24] Kinect v. Vicon MX [26] Kinect V2 v. Vicon MX [29]
		Normal gait	Step dimensions	Meters	0.02 [22]	0.05 [22]	2/2 [22]	-
		Gait, stairs, sit-to-stand	COM Speed	Meters/second	0.01 [29]	0.06 [24]	1/7 [22], 1/7 [29]	4/7 [24], 1/7 [29]
		Sit-to-stand	COM displacement	Meters	0.07 [29]	0.07 [29]	-	2/2 [29]
		Sit-to-stand	COM acceleration	Meters/second2	0.01 [29]	0.01 [29]	1/2 [29]	1/2 [29]

(MLS = MBS) = the markerless motion capture system produced a measure that was not statistically different from the marker-based motion capture system measure; * = number of non-statistically different results over the total number of results identified across all studies for that physical movement measure (MLS ≠ MBS) = the markerless motion capture system produced a measure that was statistically different from the marker-based motion capture system measure; ** = number of statistically different results over the total number of results identified across all studies for that physical movement measure; ROM = range of motion; COM = center of mass; - = no data available for this outcome.

summarizes the movements and measures comparing markerless and marker-based systems. We have grouped these results in into upper, lower, and whole-body measures to best reflect clinical applicability. The specific systems compared in each study are listed; most systems were ambivalent in their comparison of measurable mean differences, except for the two RGB cameras paired with a Motion Analysis Corporation system, which were only dissimilar from one another in one of the eight total measurements [19]. Conversely, five of six measurements were dissimilar in the comparison between four RGB cameras and a Vicon MX system [17].

5.3. Upper Body

Five studies compared upper body joint-specific measures during movements including functional reaching [16], side bending [17], jump landing [20], squats [20], lifting [21], and walking [25,28]. From these seven studies, 29 measures were identified and are summarized in Table 2. Although a total of 13 measures were found to be statistically different between measurements by the two systems, more evidence was found supporting no between-system differences during assessment of joint ROM displacement for shoulder flexion and extension [25], shoulder abduction and adduction [21], trunk flexion and extension, as well as peak joint angle for shoulder elevation and depression [25] and trunk flexion and extension. However, several instances of evidence were found indicating that there were significant differences in the evaluation of peak joint angle for shoulder flexion and extension [25], as well as joint ROM displacement and peak joint angle for elbow flexion and extension [25]. The largest mean difference detected between measurements from the two systems was 29.01 degrees in peak elbow flexion/extension joint angle [25]. With respect to trunk side flexion [17], a similar incidence of significant and non-significant differences in between-system measures yielded inconclusive evidence for the assessment of this movement. Overall, upper body measures during physical movements were heterogeneous and conflicting across studies, with more evidence in favor of MLSs producing similar outcomes to MBSs for the measures related to joint displacements in shoulder flexion/extension, shoulder elevation/depression, and trunk flexion/extension movements.

5.4. Lower Body

Twelve studies compared joint-specific measures during lower body-related movements including jump landing and squats [18,20,27], lifting tasks [21], normal gait [22,26,28,32], running [23], stair navigation [18,24], and upper extremity reaching tests [30]. In these studies, 225 different measures were recognized and are outlined in Table 2. Overall evidence indicating significant differences between the two systems was found to be more prevalent, including for the assessment of peak joint angle for hip abduction and adduction [20,31], knee valgus and varus [18], and knee flexion and extension [18,23,24,26,31], as well as for both the peak joint angle and joint ROM displacement in ankle dorsiflexion and plantarflexion [22,24,28,30,31]. The largest mean difference of between-system outcomes was 21.17 degrees when measuring joint ROM displacement of the ankle in plantarflexion/dorsiflexion [24]. Similar measures between MLSs and MBSs were found only for the assessment of joint displacement of the hip in abduction and adduction [16,17,18,19,23,24], the hip in flexion and extension [20,21,22,24,26,31], the knee in valgus and varus [20,27], and for pelvic tilt [22]. Conflicting evidence was found for the assessment of the peak joint angle of the hip in flexion and extension, as well as for the total joint displacement of the knee in flexion and extension, which precluded drawing a conclusion of MLSs being comparable to MBSs for these measures.

5.5. Whole-Body Measures

With respect to whole-body measures, four studies examined measures obtained during movements such as gait [22,26], stair navigation [24], and sitting to standing [29]. Thirty-three measures were identified and are reported in Table 2. Again, overall evidence indicating significant differences between the MLSs and MBSs was more prevalent, including stride time [24,26] and the speed of displacement and total displacement of the center of mass [24,29]. The greatest mean difference was measured during displacement of the center of mass, with a difference between means of 0.07 m [29]. There was more evidence in favor of the MLS being similar to the MBS only for the assessment of step length [22], while the overall evidence was inconclusive for the comparison between the MLS and the MBS for the assessment of acceleration of the center of mass [29]. Overall, among whole-body measures, step dimensions were the ones that presented most consistently without significant differences between systems.

5.6. Considerations for Future Clinical Research Involving MLS

For our aim of exploring the procedural characteristics of MLS technologies in measuring physical movements that should be considered when planning upcoming clinical research, we explored the population, technology, design, and setup of the studies.

5.7. Populations Assessed

Participant samples varied across the included articles. Average participant age was 29.2 (±9.3) years across the included articles, and sample sizes ranged from 1 to 78 participants. Participant age was not reported in one study [16]. Representation of each biological sex assigned at birth was roughly equal among included studies; however, two of the studies [21,28] included exclusively male participants, and one study did not report the assigned sex at birth of the participants [16]. In all but three articles [17,23,25], participants were identified as being healthy. Most articles did not define “healthy” populations, but those that did generally outlined a healthy population as: “free from musculoskeletal, neurological, or cognitive injury or impairment that would impede the performance of the movement task in the study” [24,25,26,31]. The identified conditions in non-healthy populations included lower back pain [17], patellofemoral pain syndrome [23], and acquired brain injury [25].

5.8. Characteristics of the Technology

The reviewed literature compared few characteristics of MBS and MLS technologies. MLSs used between one [16,17,18,20,24,25,26,27,29,30,32] and four [22] cameras to capture physical movements. The Microsoft Kinect V2 was used in eight studies [17,18,20,24,25,29,30,31], the Microsoft Kinect in three studies [26,27,28,31], 2D RBG cameras were used in three studies [21,22], the Azure Kinect in two studies [16,31], while a smartphone camera [23] and the Orbbec Astra [32] MLS were each used in one study. While the Kinect system has been discontinued (2017) and replaced by Azure Kinect, research and clinical practice may continue to utilize the older form of technology [9].

Table 3. Summary of technology and procedural characteristics of the included studies.

Study [Ref]	Type of MLS	MLS Software Used	Number of MLS Cameras	Type of MBS	Number of MBS Cameras	Number of MBS Markers	Additional Equipment Used	Camera Positioning and Fixation
[16]	Azure Kinect	Custom SDK	1	Optitrak	-	-	Serious game system	Fixed at 0 degrees
[17]	Microsoft Kinect V2	Qinematic v2.1.20	1	Tracklab BTS-elite	8	15	Kistler force plate	Fixed at 0 degrees 3 m from participant
[18]	Microsoft Kinect V2	DLL.NET	1	Vicon MX	8	16	Tripod	Fixed
[19]	4 × 2D RGB Camera	OpenPose (2022)	4	Vicon MX	13	12	-	-
[20]	Microsoft Kinect V2	PhysiMax v2.11	1	Vicon Bonita	10	45	Force platform	Fixed 3.4 m from participant
[21]	2 × 2D RGB cameras	OpenSim (2018)	2	Motion Analysis Corp	-	45	-	Fixed at 90 degrees and 135 degrees
[22]	3 × 2D RGB cameras	AdaFuse (2021)	3	Optitrak	8	22	Tripods x3	Fixed at 0, 45, and 315 degrees
[23]	2 × 2D high framerate iPhone cameras	Custom MATLAB (2015)	2	Codamotion	4	24	Force plate; tripods x2	Fixed at 90 degrees 2.5 m from participant, and 0 degrees 6.5 m from participant
[24]	Microsoft Kinect V2	DLL.NET	1	BTS infrared	8	16	3-stair platform	Fixed 2.5 m from participant
[25]	Microsoft Kinect V2	« Standard » Kinect SDK	1	Optitrak	13	65	-	Fixed
[26]	Microsoft Kinect	Brekel Kinect	1	Vicon MX	10	29	Treadmill	Fixed at 315 degrees
[27]	Microsoft Kinect	Kinect SDK v1.8	1	Motion Analysis Corp	10	28	Goniometer; metronome	Fixed at 0 degrees 1–3 m
[28]	Microsoft Kinect	iPi Recorder v2.2.2.27	2	Light infrared high-speed cameras (Oqus 300 series)	8	33	Force plates x3	from participant
[29]	Microsoft Kinect V2	Kinect SDK v2	1	Vicon MX	7	33	Chair; tripod	Fixed at 43 and 317 degrees 3.4 m from participant
[30]	Microsoft Kinect V2	Kinect SDK v2	1	Vicon MX	8	33	Tripod	Fixed at 0 degrees 3 m from participant
[31]	Azure Kinect; Microsoft Kinect V2; Orbbec Astra	Azure Kinect body tracking-SDK-v0.9.4; Kinect SDK v2; Orbec SDK v2.0.17	1	Vicon Bonita	12	16	Leveled treadmill; tripods x3	Fixed at 0 degrees 2 m from participant

[ref] = study reference; MBS = marker-based motion capture systems; MLS = markerless motion capture systems; - = no data for this measure from the study; m = meters; 2D = two-dimensional; DLL.NET = dynamic link library with .NET framework and customized MATLAB code; RGB = red, green, blue.

highlights the types of MLSs and cameras used. For MBSs, there was a wider variety of systems used. Seven research groups chose to use Vicon systems [18,20,26,29,30,31], three studies used Optitrak [16,22,25], while BTS [17,24] and Motion Analysis Corporation [21,27] systems were each used twice. Moreover, Codamotion [23] and Oqus [28] systems were each used only once. On average, 28.8 (±14.4) markers were fitted to participants being measured by an average of 9.1 (±2.3) cameras, with a range of four [23] to 13 [25] cameras used.

The included studies analyzed the data collected by MLS using the following software: Microsoft Kinect SDK v2 [29,30,31], “Standard Microsoft SDK” [25], a skeleton model of Brekel Kinect Software [26], Microsoft SDK v1.8 [27], Azure Kinect Body Tracking-SDK-v0.9.4 [31], Dynamic Link Library with .NET framework and custom MATLAB code [18,24], OpenPose [19], OpenSim [21], Custom SDK [16], Custom MATLAB program (2015) [23], Qinematic v2.1.20 [17], PhysiMax Technologies Ltd. Bronx, NY, USA proprietary software v2.11 [20], AdaFuse (2021) [22], iPi Recorder v2.2.2.27 [28], and Orbec SDK v2.0.17 [31]. The software used in each study is summarized in Table 3.

5.9. Experimental Setup and Procedures

The experimental setups were described with varying detail across the 16 studies. Inclusion of alternate equipment was observed in 13 studies [16,17,18,20,22,23,24,26,27,28,29,30,31]. To complement MBS measurements, force plates were used in four studies [17,20,23,24,28] to assess ground reaction force during gait, and a goniometer was used once to evaluate joint angle range of motion [27]. Other equipment such as a treadmill [26,31], a chair [29], and a three-stair platform [24] were used to facilitate desired movements. In terms of movement instruction, no studies explicitly commented on how the participants were cued to carry out the movement and none included rehabilitation professionals in their in-person procedures. With respect to issues with experimental setup and procedures, the included studies reported participant movements perpendicular to camera lenses [18], self-occlusion by participants [21], and blurred motion during rapid movements [22] as barriers to MLS function. MLS use entails a straightforward setup, a simple camera (in some contexts, more than one), flexible space requirements, and the motion capture process. Since all studies were conducted in person in a lab or a motion-capture studio setting, it is not possible to comment on MLS procedural or technical translation to clinical or remote contexts. Furthermore, none of the included studies incorporated intentional error testing in their experimental setup or procedures. Lastly, none of the studies reviewed for our scoping review were designed to assess nor explain current or future usability of MLSs in clinical contexts, and therefore, we could not evaluate application.

6. Discussion

Within the 16 included studies, nearly two-thirds of statistical tests produced p values indicating that the physical movement measures captured by MLSs were statistically different from those captured by MBSs, suggesting that MBSs and MLSs are not yet interchangeable. There were no trends between specific MLSs and MBSs, but the overall results showed greater differences in the assessment of lower body movement measures between the two systems compared with upper body movements, indicating that there may be more differences between the two types of system when the targeted measure is closer to the ground. To explain such findings, some have hypothesized that end-range joint angles and self-occlusion may be responsible for increased discrepancies between MLSs and MBSs in assessing the lower body [21]. From a clinical perspective, placement and use of MLSs would need to be taken into consideration to apply these technologies to patient groups known to have mobility and balance issues. For example, measures of gait, transfer, and balance could be more effectively accomplished with a portable system; however, the potential for differences in movement outcomes would need to be taken into account.

Despite all studies [16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31] disclosing parameters with statistical significance, only some articles mentioned clinically relevant differences [17,29,30], while none set explicit parameters for clinically relevant differences for their tests. Future studies would benefit from discussion of clinically relevant differences and setting parameters for the same [10,32]. Regarding the direct application of MLSs, although the potential of MLSs was highlighted as a more accessible, user-friendly option compared with MBSs, none of the studies reviewed were designed to focus on direct application. We have highlighted three areas that future research could target when considering application: contextual design; consideration of intentional error testing; and assessment of patient populations.

6.1. Contextual Design

The studies we reviewed were all conducted in laboratory or studio settings without details surrounding instruction for movement. It is anticipated that studies exploring and commenting on virtual and remote use of motion capture systems will eventually be designed to focus on remote accessibility of the technology, such as designs that do not involve direct assessor engagement with participants or that use virtual settings. Furthermore, independent user setup could be a means of assessing how MLSs can be used for telehealth purposes, where the technology expert may not be in the same space as the user. Basic simulation and clinically contextual studies have shown promising results relating to how such interventions can translate to healthcare settings [33,34].

6.2. Intentional Error Testing

The included literature captured the motion of healthy participants while carrying out ideal movement patterns and assessed outcomes based primarily on degrees of ROM. Many of the included studies recommended MLSs as potential alternatives to MBSs under these conditions [18,20,21,22,24,27,28,29,30,31], but certain studies acknowledged the need for future research focused on clinical relevance, i.e., larger sample sizes, measuring impaired movement patterns, and less than optimal environmental conditions [18,20,24,25,26,27,28,31]. This acknowledgement is relevant because in healthcare contexts, movement testing typically includes participants who are unable to carry out the physical task in the ideal way and often includes conditions that are imperfect [33]. None of the studies explored less-than-ideal movements nor intentional error testing (i.e., creating awkward movement patterns, incorrect movements, or incomplete movements) to see how the systems would compare in relation to more clinically prevalent movements. Additionally, the setup for evaluation across the studies was aimed at maximal movement capture, such that less-than-ideal settings were not tested (i.e., poor lighting, obstructive clothing, or non-optimal MLS camera setups); none of the included studies were designed to assess joint angles skewed by intentional errors. In future studies, testing that includes poor lighting, loose clothing, and common error patterns could be considered in order to obtain evidence of the usability of MLSs in context.

6.3. Assessment of Clinical Populations

Across the studies, study participants had an average age of 29.2 (±9.3) years. The populations that are driving an increase in demand for rehabilitation services are typically above the age of 65 [35]. Individuals in this older age group are known to have higher occurrences of physical limitations compared with the younger participants in the included studies. There is limited evidence to support the consistent accuracy of MLSs across varying age groups. The contrast in age between the populations most frequently accessing rehabilitation services and the participants of the reviewed studies suggests that a more clinically representative study sample would offer more clinically translatable results. Furthermore, populations with complex chronic conditions are also one of the key driving factors in the increased demand for rehabilitation services [36]; in the included studies, only 19.8% of participants were observed while living with pain or chronic disease [17,23,25].

6.4. Strengths

Our search included major databases across fields of both healthcare and technology; included studies were broad, to avoid restricting or limiting literature, with no restriction on journal, originating country, nor study context. Our review is practical as it includes a general population of adults (as opposed to only elite athlete participants). By limiting our findings to direct comparisons, we were able to isolate mean differences across the two types of system and directly compare findings, making the results relevant in determining whether a system could be used for physical measures. Our review was conducted using Arksey and O’Malley’s established framework, showing rigor and transparency [12,13].

6.5. Limitations

By limiting our findings to peer-reviewed published journals, we may have missed newer technologies in motion capture not yet published in this form. By limiting studies to those including direct comparisons, we were unable to comment on the body of literature which evaluated correlations between the two techniques. While statistical differences have been reviewed and described as part of this review, it would be helpful in future studies to add consideration of testing for clinically relevant differences, validity, and reliability. Furthermore, we may have encountered publication bias in the studies that were completed and published being conducted in laboratory settings rather than in technological or commercial environments. Changes in technology, such as technology becoming obsolete during the course of our review, reflect the limitation of the speed of thorough research compared with the emergence and retirement of technologies.

6.6. Relevance

This scoping review first explored comparison of physical movement measurements in adult populations captured by MLSs and MBSs and then explored the considerations for application of MLSs. This review supports the need for contextual use of technology as there are differences in measures between MBSs and MLSs, particularly for upper and lower body movements and peak joint angles. Selection of technology for motion capture in rehabilitation settings should also consider cost, size, efficiency, accessibility, and relevant outcome and assessment potential for use across healthcare services.

6.7. Future Research Implications

This review identifies the need for future validation of MLSs in comparison to MBSs for rehabilitation healthcare contexts. This review also identifies the use of translatable research designs, including the use of patient populations and testing with standardized clinical outcomes and tools. Moving forward, these strategic approaches to evaluation, when applied to future healthcare research, can facilitate viable motion capture studies for healthcare and rehabilitation.

7. Conclusions

While there is potential for MLSs to be used for motion capture, the current differences found between MLSs and MBSs remain consistent across studies and movements, particularly those of the lower body. While promising, the application of MLSs in research is still in development. Future studies can be conducted from the perspective of application and use, with consideration of the populations that are evaluated and the common movement errors they make, the types of movements that are tested, and the context in which the testing is performed.