Simultaneous Validity and Intra-Test Reliability of Joint Angle Measurement through Novel Multi-RGB Sensor-Based Three-Joint-Continuous-Motion Analysis: A Pilot Study

Abstract

The use of motion-analysis devices that can measure the progress of rehabilitation exercises for nerve paralysis is increasing because of the need to confirm the effectiveness of treatment for sports injuries. This study developed a new motion-analysis device that can be easily handled compared with the existing VICON motion-analysis device. Motion analysis of the human body (specifically, hip flexion, knee flexion, and trunk rotation) performed simultaneously with the new device and the existing VICON device was compared. Five healthy young men voluntarily participated in this study. Various joint angles were captured using a marker-less multi-view image-based motion-analysis system and a VICON motion capture system with markers during lower-extremity work. Intra-class correlation coefficient (ICC) analysis was used to examine simultaneous- and angular-limit validity and the intra-joint reliability of multi-point image-based motion-analysis systems. Simultaneous validity analysis showed that the highest ICCs for hip flexion, knee flexion, and trunk rotation were 0.924–0.998, 0.842–0.989 or higher, and 0.795–0.962, respectively. We confirmed that this new marker-less motion-analysis system has high accuracy and reliability in measuring joint kinematics in the lower extremities during rehabilitation and in monitoring the performance of athletes in training facilities.

1. Introduction

The existing patient-specific treatment device field is not as active as expected because of insufficiencies in the service ecosystem and structural limitations in the service operation at the independent application level. Therefore, it is necessary to combine biotechnology with information and communication technology (ICT), both of which are strong technologies in South Korea, to establish an ecosystem in the field of patient-specific treatment devices, such as the newly developed motion-analysis system. Figure 1 shows that the global market for digital therapy reached USD 3.23 billion in 2021 from USD 2.7 billion in 2020. It is expected to reach USD 4.66 billion by 2023 and USD 17.34 billion by 2030. The compound annual growth rate (CAGR) for the digital therapy market is 20.5%, which is significantly higher than that of the general pharmaceutical market. However, the digital therapy market has different forecasts for each institution. As a result, there are limitations in predicting the exact market size or growth rate. This technology, which combines information technology with existing medical and pharmaceutical technologies, is considered an alternative tool for overcoming the limitations of existing medical treatments; therefore, its market growth rate is significantly higher than that of the existing medical industry market [1]. In particular, the patient-specific treatment device market is expected to reach approximately USD 4.2 billion by 2021, with a CAGR of 26.1% from 2022 to 2030, because of increased demand for cost-effective digital health technology, integrated medical systems, and patient-centered care for patients in developed and developing countries. To solve medical accessibility problems such as those for degenerative brain diseases, mental diseases, and psychological (mental) disorders caused by aging, it is necessary to establish a patient-specific medical service demonstration and support system, such as a motion-analysis system. Therefore, multi-disciplinary research and development projects are needed to demonstrate patient-specific treatment devices. This is an urgent time for international exchanges because they must lay a foundation that cannot be accessed at the private level [2].

Human motion-analysis technology can be expanded to various smart healthcare areas; therefore, it can be widely used for healthcare for the elderly who are living alone, emergency medical system linkage after sustaining fall injuries, smart exercise coaching, joint function self-evaluation, physical vitality index, and daily life activity level evaluation. Furthermore, owing to the development of the internet of things and ICT such as patient-specific medical services, artificial intelligence, and big data, the need to foster a patient-specific medical industry via the commercialization of patient-specific treatment devices is emerging. Quantitative measurements or evaluations of muscle strength and body composition (body fat) in the field of healthcare have focused only on physical fitness tests. However, the digital health industry, which is the mainstay of the fourth industrial revolution era, has grown rapidly. Measuring devices for movement and posture that are related to musculoskeletal dysfunction are being developed for use in a more convenient manner, and the importance of their evaluation is highly recognized. Several overseas experts are conducting different studies to analyze the asymmetric movement and posture of the human body by using the latest technologies and equipment to alleviate pain and reduce the risk of injury [3]. However, in the field of rehabilitation, most are focused on intervention and the qualitative evaluation of experts (humans). Commonly used motion-analysis systems, such as VICON systems, are highly reliable for research and analysis; however, they can be difficult to operate depending on the situation because they are expensive (particularly with regard to spatiotemporal constraints). Therefore, these systems are widely used for research purposes rather than for practical applications in the field. Existing pose-estimation technology is based on 2D or Kinect sensors; however, the results are inaccurate and difficult to use as objective indicators. The need for related research has emerged with the recent development of 3D sensing technology. Given that pose-estimation technology is not actively used in the clinical field, most preceding studies are published in the field of general engineering, and research that can present the clinical meaning, value, and utilization of the data extracted by this technology is urgently needed.

Recently, multi-view image-based motion-analysis systems that are capable of 3D implementation have significantly affected the research and diagnosis of the musculoskeletal system in sports players and even ordinary people. These systems have the ability to simultaneously observe and analyze various movements by using images obtained from various angles in 3D [4]. These 3D-based multi-view image sensor systems are of great help in identifying and preventing potential injury factors in sports players who may experience musculoskeletal accidents when performing high-intensity movements in competitions. The multi-dimensional data collection and processing capabilities of a multi-dimensional system allow athletes to analyze their movements accurately and quickly identify the causes and effects of injuries [5]. The validity of the angular root may indicate whether the measured angular data of the system correlates with the actual motion [6]. This validity is important for reliable motion analysis and the determination of the cause of injury in sports musculoskeletal accident research [7,8]. The reliability of the 3D implementable multi-view image-based motion-analysis system introduced in this study indicates the consistency and stability of the results when repeatedly analyzing the same movement using this system. In situations involving sports musculoskeletal accidents, this system can facilitate a better understanding of accident-related factors and the development of safe exercise methods [8].

The meaning, advantages, and disadvantages of motion capture are well known. Marker-based motion capture technology is widely used for motion analysis; it involves the capture and analysis of data by using cameras and small markers to track a person’s musculoskeletal movement and estimate the actual joint position [9]. Marker-based motion capture allows the precise tracking of musculoskeletal motion, thereby enabling accurate motion analysis [10]. Motion analysis using this motion capture device is well-established and is widely used in research and various industries [11]. Additionally, it is compatible with various motion-analysis systems, can be easily adapted for multiple purposes [11], and can provide instantaneous feedback during movement by allowing for real-time tracking [12]. However, marker-based systems have several disadvantages, and the special cameras and markers used in these systems are expensive. In addition, the marker may be obstructed from the field of view of the camera used, thus resulting in incorrect or missing motion-tracking data [13].

This pilot study aimed to develop a new multi-view image sensor-based system (namely, 4DEYE) to compare and analyze the reliability and validity of the sensor developed via the simultaneous analysis of three joint angles with existing VICON sensors (VICON motion capture system). In addition, we present the possibility of the clinical application of a new motion-analysis system that has the advantages of convenience and economic feasibility.

2. Materials and Methods

2.1. Participants

Five volunteers were selected to participate in the experiment, and the conditions were set at 25.4 ± 2.0 years old, 174.4 ± 5.0 cm height, and 68.9 ± 6.8 kg weight. These volunteers participated in the pilot study for the accumulation measurement of joints where multiple joint movements occur continuously.

However, people with a history of neurological, musculoskeletal, or cognitive system disorders were excluded from this study. Before starting this study, all participants were required to hear an explanation of the purpose and process of this study and sign and submit a consent form. The experimental schedule thoroughly followed the principles of the Helsinki Declaration [14], and this study was conducted with the approval of the institutional review committee of a specialized institution. Five RGB cameras (4DEYE, SYM Healthcare Inc., Seoul, Republic of Korea) were used for multi-view image-based motion analysis in this study. In addition, data were collected by photographing participants’ postures at 12 Hz in four different directions using the configured multi-view image collection system (Figure 1). After collecting images, the angles of the hip and knee joints were analyzed using an open-source customization analysis program.

2.2. Measurement

Since it is difficult to accurately determine the operation timing of the 4DEYE and VICON systems, the minimum and maximum areas of each data point were analyzed at the same time and compared with each other. In addition, after selecting one representative indicator of 4DEYE and VICON, the minimum and maximum areas of the indicator were applied to all indicators. Up-sampling was performed to match the same 100 frames of 4DEYE and VICON using MATLAB because of the difference in the number of image frames between 4DEYE and VICON. The correlation between the two data types was obtained and the data were analyzed according to the correlation with the graph image; the measurement process is shown in Figure 2a.

The analysis methods used for the 4DEYE and VICON systems were slightly different, resulting in some differences in angle values. This is because the operation time of 4DEYE and VICON must be set manually, not automatically, resulting in experimental errors. After the VICON device was activated, the processed data was standardized to have 100 frames. However, the data that were not processed after operation showed some significant differences in angle value compared with 4DEYE. The images for matching VICON and 4DEYE are shown in Figure 2b,c.

Hip flexion measures the angle of the hip-joint vector (the line connecting the hips and knees) compared to the upper-body-center line (the line connecting the center of both shoulders and the center of both hip joints on the z-axis). The picture above shows the right side as an example, so for convenience, it is assumed that the upper-body-center line has been moved parallel to the right. Therefore, hip flexion measures the angle of the thigh vector (line connecting numbers 18 and 19 in Figure 2d) compared to the upper-body-center line moved parallel to the right. Knee flexion measures the angle between the thigh vector (the line connecting numbers 18 and 19 in Figure 2e) and the calf vector (the line connecting 19 and 20 in the picture). Trunk rotation measures the angle between the y-axis (left and right reference lines) and the straight line connecting both shoulders (line connecting numbers 6 and 12 in Figure 2f) on the horizontal plane (x–y plane).

Currently, the mark-based VICON system is widely used for motion analysis (ex, clinical field) and is used as a standard analysis system for human motion analysis [15,16] because it does not require a separate wired device, and the weight of the marker is light. However, the marker must be artificially attached to each subject, the preparation for research is complicated, and there are many spatial constraints because the marker is very sensitive to light [17]. In addition, the device system is expensive due to the use of expensive infrared cameras [18,19].

In general, the RGB sensor system combines a 3D reconstruction algorithm based on a deep learning multi-view 2D image to estimate a 3D posture. RGB-D sensors for quantitative evaluation or motion analysis can be a good alternative to existing sensors [20], and commercial RGB sensor devices with motion analysis describes 4DEYE^®. Existing deep-learning-based posture estimation models are OpenPose made by Carnegie Mellon University and MediaPipe made by the Google company. There are two approaches: one is a bottom-up approach that first recognizes the human segment on the image and then finds and groups the main parts of each individual based on the detection results. The other is a top-down approach that first recognizes the main parts of the entire image and then constitutes the human segment for each individual. In addition, the latest computer vision technology can analyze movement by integrating and converting multiple 2D images into 3D [21].

2.2.1. VICON Motion-Tracking System

The VICON motion-tracking system (MXT series, Oxford Metrics, Ltd., Oxford, UK) consists of proprietary hardware using eight infrared (IR) cameras to coordinate location points. This system required a counter-reflection marker for the IR optical signal from the IR strobe of each camera. The torso and lower limb landmarks included the 7th cervical spine (C7), 8th thoracic spine (T8), and the xiphoid process of the jugular vein and sternum, to which four markers (14 mm in diameter) were attached. In addition, two cross-shaped clusters with four markers each were attached to the thighs and shanks, respectively, and aligned with the femur or tibia axis to analyze one axis of the cross. Each camera used in the experiments for this study photographed the 3-dimensional posture of the marker in 100 frames, and the joint angle was calculated similar to an analysis based on a multi-view motion capture system. Using the positional points of each marker, we set the body coordinates according to the existing research methods, and we set the femoral- and tibia-axis vectors [22]. In addition, cross-shaped clusters were used for vectors of the femur and tibia axes, and joint angles were analyzed using MATLAB R2018A (MathWorks, Inc., Natick, MA, USA).

2.2.2. Multi-View RGB Senor-Based Motion-Analysis System (4DEYE)

Utilizing a multi-view RGB sensor-based motion-analysis system (4DEYE, SYM Healthcare Inc., Seoul, Republic of Korea), which consists of five cameras, the subject’s motion was photographed in 12 Hz areas in five different directions. In addition, a custom OpenCV- [23] and OpenPose [24]-based analysis program was used to analyze hip and knee joint angles from images of movements in various directions. In detail, the OpenPose software (https://cmu-perceptual-computing-lab.github.io/openpose/web/html/doc/md_doc_00_index.html accessed on 20 November 2023) can acquire five images simultaneously using five cameras, i.e., estimating the 2D position of seven physical key points: neck, left shoulder, right shoulder, right hip, knee, and ankle. Then, based on information on the relative position and direction of the camera, the 3D positions of each key point were re-implemented in four 2D positions corresponding to the key point using OpenPose software. The bending and stretching of the hip joint were analyzed indirectly as the angle of the thigh axis for the entire body, and the bending or stretching of the knee was analyzed indirectly as the angle between the thigh and the tibia axis. Body coordinates are analyzed by defining the z-axis as a vector from the joint to the neck, and the x-axis as a vector perpendicular to a plane formed by the left shoulder and right shoulder.

In addition, x- and y-axes were defined as vectors orthogonal to the x- and z-axes, and the femur and tibia-axis vectors were defined as vectors pointing from the right hip to the knee and from knee to the ankle, respectively. The hip-flexion angle was calculated as the angle between the negative z-axis and thigh-axis vectors of the body coordinates, and hip flexion was quantified in the 3-dimensional motion regardless of the sagittal plane. The hip-flexion angle to lift the leg increased from 0° (i.e., anatomically neutral position) to 120°. Finally, the analyzed joint angle was calibrated to match the length with the data collected in the 100 Hz area using the VICON motion-tracking system. Detailed equipment specifications for VICON^® and 4DEYE^® are attached in

Table 1. Equipment specifications.

Sensor	RGB Sensor	Infrared (IR) Sensor
Camera images
Camera type	RGB camera	Vicon camera
Device model	Joytron-HD20	Vicon T10
Resolution	1920 × 1080	1120 × 896
Frame	30 fps	250 fps
Support system	Windows xp, Vista 7,8/Mac OSx10.4.8	Nexus 2.x, Shogun 1.x

.

2.3. Procedure

The experiment required performing a high-level continuous motion called hip rock, which combines hip-flexion, knee-flexion, and trunk-rotation movements by using multiple joints. The experimenter explained the motion to the subject, who was trained in the movement before performing it. Starting with the standing operation, the hip rock operation was performed to minimize trunk rotation, and the upper and lower limbs were twisted while maintaining 90° hip flexion and 90° knee flexion. The motion was repeated five times, and the two capture systems were simultaneously activated when the subject started and completed the motion while maintaining the final posture. The break interval was 3 to 5 min. A researcher with a physical therapist license conducted the inspection and management directly to secure the subject’s performance ability and to maintain the accuracy and best condition of the motion. Joint angle data for hip and knee flexion were collected and processed during the lower-extremity task, and additional analysis was conducted using data from each attempt and the data from the average of attempts (Figure 3).

2.4. Statistical Analysis

Statistical analysis based on the results of this study included basic mean and standard deviation analysis. Simultaneous and angle-limited validity (intra-class correlation coefficients [ICC] 3, k) analysis with the new multi-view image-based motion-analysis system (marker-less) developed in this study was conducted and compared with the VICON motion capture system (including markers). ICC analysis used ICCs and 95% confidence intervals (CIs) to evaluate the in-screen reliability (ICC3, k) of each motion-analysis system. ICC values were interpreted as follows: defect, ICC < 0.50; defect, 0.50–0.75; good, 0.76–0.90; and high, 0.90. The significance level was set at p < 0.000. All statistical analyses were performed using SPSS for Windows (version 18.0; SPSS Inc., Chicago, IL, USA) or Microsoft Excel 2019 (Microsoft Inc., Redmond, WA, USA).

3. Results and Discussion

This section is divided into several subsections. A concise and precise description of the experimental results, their interpretation, and experimental conclusions are provided. The x-axis was a frame, the maximum and minimum sections were cut, and the frames in the section were matched at 100 frames using VICON and 4DEYE so that the result and graphs could be produced. Subsequently, the interclass correlation coefficient was obtained from the statistical analysis of the results. The y-axis represents the angles of hip flexion, knee flexion, and trunk rotation. The angle of the body was obtained by dividing the frame of the sample motion by a constant time. VICON and 4DEYE observed that the degree increased as the number of frames increased over time. This result was the same in the subsequent knee-flexion and trunk-rotation analyses, as for hip flexion.

As a result of comparing motion analysis using IMU sensors and OpenPose-based RGB sensors in previous studies, the kinematic accuracy between the two equipment was consistent as shown in this study [25].

The three reasons for the differences in angle on the graphs are as follows.

First, when analyzing the operation with VICON^® an error may occur from the attachment of an infrared reflection marker because of skin flexion. The length of each segment was measured for the length of the femur in the direction connecting the knee from the hip, and for the length of the fibula in the direction connecting the knee to the ankle from the knee, to measure the joint angle of the knee joint. The tension of the skin changed as the body moved, resulting in bending, and the line connecting the two markers did not match the longitudinal direction of the body when the markers were attached to two points starting from the bones protruding into the skin surface in the longitudinal direction of this segment. In addition, the clinical field stipulates that movement occurs because of the fixed axis of the joint when describing the movement of the human body; however, since there is no fixed axis at the center of the joint, the direction of movement may continue to change when actual joint movement occurs. This could lead to an error of several degrees.

Second, although the trends in statistical and graph angles were very consistent, 4DEYE shows a smoother graph compared to VICON because it is inferred by calibrating values through AI-learned algorithms in OpenCV.

Third, the VICON^® measurement requires a total of four markers to be attached up and down on the chest and back within the sagittal plane of the body (specific attachment positions are up/down of the sternum, protrusions of C7 and T8) in the case of measuring the angle of hip flexion using the difference in the estimation method of the body coordinate system. The directions of the mediolateral, superior-inferior, and anteroposterior axes can be estimated from these four markers. The direction of the body’s reference coordinate system is calculated based on the position of the center of the joint to estimate the position of the neck, shoulders, and hip joints within the frontal plane if measuring using 4DEYE^®. Therefore, the estimated reference coordinate systems of the torso, hip, and knee are different; thus, errors in the vertical degrees can occur as VICON^® and 4DEYE^® have different reference points for estimating the coordinate system. Additionally, if this error is added to the error caused by the first cause described above, it can result in larger errors without offsetting each other.

3.1. Analysis of Hip Flexion

Figure 4 shows the degree of hip flexion using a comparative analysis of VICON (black) and 4DEYE (red). The hip-flexion analysis graph shows the increase in degree according to the overall increase in the frame for each subject. In the left-side analyses for Subjects 2 and 3, the result graphs of the 4DEYE and VICON sensors were almost identical; for Subject 1, the sensors had similar degrees throughout the frame. In the second half of the frame, the 4DEYE sensor exhibited a slightly higher degree than the VICON sensor. For Subject 4, the VICON sensor had a slightly higher degree than the 4DEYE sensor in the overall frame section. In the hip-flexion analysis, the resulting graphs of the 4DEYE and VICON sensors were almost identical in the case of Subject 2; for Subjects 3 and 4, the rate of degree–increase according to the frame was slightly higher than that of the 4DEYE sensor. Subject 1 predicted that the 4DEYE sensor results would show a sudden decrease in degree at a high frame level owing to various errors during the experiment.

3.2. Analysis of Knee Flexion

The degree of hip flexion was compared between VICON (black) and 4DEYE (red) results using comparative analysis (Figure 5). In the hip–knee analysis graph, similar to the previous hip-flexion analysis, the degree did not increase as the frame increased, but it appeared in many subjects. In the left-side analysis, all subjects except Subject 4 showed a similar degree in the frame in the first half. For Subjects 1 and 3, the 4DEYE sensor showed a higher degree than the VICON sensor in the second half of the frame; however, for Subject 2, the 4DEYE sensor showed a lower degree than the VICON sensor. The VICON results for Subject 4 were comparably different from those of the 4DEYE sensor. The hip–knee analysis on the right side showed a tendency for the degree to increase as the frame increased, that is, a tendency that is somewhat different from the previous graph trend (Subjects 1 and 4). For Subject 2, the graphs for the 4DEYE and VICON sensors were almost identical. For Subject 3, the VICON sensor obtained a higher degree in the early frame than the 4DEYE sensor; in the second half, the 4DEYE sensor obtained a higher degree than the VICON sensor.

3.3. Analysis of Trunk Rotation

Figure 6 shows the degree of trunk rotation via a comparative analysis of VICON (black) and 4DEYE (red). In the trunk-rotation analysis, as the frame size increased in many subjects, the difference in the rate of increase between the VICON and 4DEYE sensors was significant. In the analysis on the left, the degree tended to increase to negative values as the number of frames increased.

In the VICON system, rotation can be seen as a trunk z indicator; in the 4DEYE system, the negative (−) value is set to increase toward the left, and the resulting value of 4DEYE is converted according to the setting of VICON. Therefore, when the trunk rotates to the left, the angle becomes more negative. Looking at the trend, the 4DEYE sensor had a higher negative increase rate of degree than the VICON sensor for all subjects, except Subject 1. In the left-side analysis of trunk rotation, similar to the previous graph, the degree tended to increase according to the frames. In addition, Subjects 1 and 2 showed different patterns of degree–increase according to the frames. For Subject 1, the 4DEYE sensor showed the opposite degree according to the frame compared with the VICON sensor. For Subjects 3 and 4, 4DEYE had a high degree in the first half of the frame, but the result was reversed in the second half.

3.4. Statistical Analysis

Table 2. Validity of the angle compared between the novel 4DEYE system and the VICON motion capture system.

Movement		Left				Right
Body Regions	Subject	M	SD	ICC(3, k)	95% CI	M	SD	ICC(3, k)	95% CI
Hip Flex.	1	94.02	43.15	0.958	0.938–0.972	108.71	41.88	0.982	0.973–0.988
	2	80.38	29.44	0.998	0.997–0.999	98.38	33.15	0.997	0.995–0.998
	3	94.4	42.36	0.972	0.959–0.981	100.57	47.25	0.997	0.995–0.998
	4	85.49	27.52	0.987	0.981–0.991	94.97	27.59	0.973	0.959–0.982
Knee Flex.	1	130.35	41.52	0.940	0.911–0.960	141.67	43.64	0.962	0.944–0.975
	2	104.57	25.06	0.997	0.995–0.998	134.09	45.37	0.997	0.996–0.998
	3	123.97	46.72	0.960	0.941–0.973	142.74	46.94	0.975	0.962–0.983
	4	133.78	49.99	0.925	0.888–0.949	140.34	35.16	0.841	0.763–0.893
Trunk Rotation	1	47.91	17.61	0.979	0.968–0.986	–37.78	17.34	0.992	0.989–0.995
	2	33.45	8.75	0.982	0.974–0.988	–45.95	11.76	0.832	0.751–0.887
	3	29.07	10.59	0.909	0.865–0.939	–26.85	10.44	0.946	0.919–0.963
	4	24.06	8.83	0.962	0.943–0.974	–65.86	20.92	0.958	0.938–0.972

CI, confidence interval; ICC, intra-class correlation coefficient.

shows the statistical values for the relationship between the 4DEYE sensor results for the three measurement targets and those from the VICON sensor. The two-way batch randomization effect model was analyzed using the ICC of SPSS v21.0 to analyze the ICC of the data. The average ICC of the average measure was analyzed, with a single measure of the ICC showing the difference between them. The ICC results were between zero and one, and the interpretation of each range was similar to that of the general correlation coefficient indicator. The ICC results are considered not good below 0.4, normal at 0.4–0.6, good at 0.6–0.75, and very good at 0.75–1.00.

Individually, Subject 1 showed an ICC of 0.982, a 95% CI of 0.973–0.988, and p = 0.000 on the right side of the hip flexion angle and an ICC of 0.958, a 95% CI of 0.938–0.972, and p = 0.000 on the left side. Subject 2 showed an ICC of 0.997, a 95% CI of 0.995–0.998, and p = 0.000 on the right side of the hip flexion angle and an ICC of 0.998, a 95% CI of 0.997–0.999, and p = 0.000 on the left side. Subject 3 showed an ICC of 0.997, 95% CI of 0.995–0.998, and p = 0.000 on the right side of the hip flexion angle and an ICC of 0.972, a 95% CI of 0.959–0.981, and p = 0.000 on the left side. Subject 4 showed an ICC of 0.973, a 95% CI of 0.959–0.982, and p = 0.000 on the right side of the hip flexion angle and an ICC of 0.973, a 95% CI of 0.959–0.982, and p = 0.000 on the left side. In the hip–knee analysis, it was found that the angle of the target porcelain was analyzed, and all targets showed a high consistency and reliability of 0.84 or higher. Subject 1 showed an ICC of 0.962, a 95% CI of 0.944–0.975, and p = 0.000 on the right side of the hip–knee angle and an ICC of 0.940, a 95% CI of 0.911–0.960, and p = 0.000 on the left side. Subject 2 showed an ICC of 0.997, a 95% CI of 0.996–0.998, and p = 0.000 on the right side of the hip–knee angle and an ICC of 0.997, a 95% CI of 0.995–0.998, and p = 0.000. Subject 3 showed an ICC was 0.975, the 95% CI was 0.962–0.983, and p = 0.000 on the right side of the hip–knee angle and an ICC was 0.960, the 95% CI was 0.941–0.973, and p = 0.000 on the left side. Subject 4 showed an ICC was 0.841, the 95% CI was 0.763–0.897, and p = 0.000 on the right side of the hip–knee angle and an ICC was 0.925, the 95% CI was 0.888–0.949, and p = 0.000 on the left side. Trunk -rotation analysis showed a high degree of agreement of 0.83 or more in the right and left movements as a value for angle change. Subject 1 showed an ICC of 0.992, a 95% CI of 0.989–0.995, and p = 0.000 on the right side of trunk-rotation angle and an ICC of 0.979, a 95% CI of 0.968–0.986, and p = 0.000 on the left side. Subject 2 showed an ICC of 0.832, a 95% CI of 0.751–0.887, p = 0.000 on the right side of trunk-rotation angle and an ICC of 0.982, 95% CI of 0.974–0.988, and p = 0.000 on the left side. Subject 3 showed an ICC was 0.946, the 95% CI was 0.919–0.963, and p = 0.000 on the right side of the trunk-rotation angle and an ICC was 0.909, the 95% CI was 0.865–0.939, p = 0.000 on the left side. Subject 4 showed an ICC of 0.962, the 95% CI was 0.943–0.974, and p = 0.000 on the right side of trunk-rotation angle and an ICC was 0.962, the 95% CI was 0.943–0.974, p = 0.000 on the left side.

The joint angle (trajectory) for complex movements in which multiple joints are used continuously should be accurately analyzed to identify dysfunction in a variety of actual movements. In addition, this study conducted a comparative analysis of the simultaneous validity of joint angle (trajectory) and reliability in the test through 4DEYE^® based on a novel multi-view RGB sensor and VICON^® with infrared-based marker attachment, which consists of hip flexion and knee flexion in the sagittal plane and trunk rotation in the transverse plane as one continuous motion as data on transverse plane movement, which could not be obtained in previous studies. Therefore, high reliability could be obtained for the joint measurement of the movement of all joints.

In the field of rehabilitation science, identifying and improving joint movement damage and mobility are important clinical evaluation points. In this study, the hip rock motion was selected to measure the continuity of multi-joint composite movements and it involves the complex movements of knee flexion, hip flexion, and trunk rotation. The in-depth relationship between the VICON and 4DEYE results is significant for moving away from marker-attached products, which are uncomfortable, when assessing joint movement damage or mobility evaluation and improvement. Motion-analysis equipment that is based on a non-marker and non-wearable multi-point image sensor can be used in various fields to improve the quality of human mobility and joint movement and can be systematically managed via quantification. In the case of trunk rotation, the existing pose-estimation engine cannot be used to target the pelvis; therefore, it may produce less accurate results than when it is used in other joints for inferring and calculating angles by using the indicators of other joints.

4. Conclusions

Simultaneous validity analysis showed that the highest ICCs for hip flexion, knee flexion, and trunk rotation were 0.938–0.999, 0.763–0.998 or higher, and 0.751–0.995, respectively. This shows a high level of reliability for this novel marker-less motion-analysis system compared with a marker-attached motion-analysis system with traditional reliability. We found that this novel marker-less motion-analysis system enables very accurate joint measurement and monitoring even when simultaneously performing combined motions with multiple planes of motion. Therefore, this system can detect damage to joint movement as an objective indicator in rehabilitation and sports, which requires precise measurements of the motion and movement of the lower extremities. Furthermore, it has accuracy and reliability for monitoring the performance of real-world interventions. Comparing and analyzing the reliability and validity of automated evaluation using a multi-view image sensor-based motion-analysis system (4DEYE) is expected to be of great help in evaluating the most difficult field of rehabilitation. Therefore, a musculoskeletal evaluation system with high evidence-based clinical practice guideline knowledge based on future clinical research should be established, and a reference value for judgment should be secured via clinical tests on actual subjects. It is expected that bad posture and movement patterns that mainly appear in patients and workers, including the general public, can be used to analyze the triggers associated with each pattern (lack of muscle strength, lack of joint motion range, etc.), derive data on “muscle skeletal disease knowledge”, and provide hope to establish a life-cycle musculoskeletal health examination database for all ages.