Validity and sensitivity of the longitudinal asymmetry index to detect gait asymmetry using Microsoft Kinect data

doi:10.1016/j.gaitpost.2016.08.022

Gait & Posture

Volume 51, January 2017, Pages 162-168

https://doi.org/10.1016/j.gaitpost.2016.08.022 Get rights and content

Highlights

•
The Continuous Relative Phase is not useable with Kinect.
•
New longitudinal index Ilong is sensitive and validated to detect gait asymmetry.
•
The Ilong index works with Kinect placed either in front of or behind subjects.
•
Only few strides (at least 5) are needed to compute a valid Ilong.

Abstract

Gait asymmetry information is a key point in disease screening and follow-up. Constant Relative Phase (CRP) has been used to quantify within-stride asymmetry index, which requires noise-free and accurate motion capture, which is difficult to obtain in clinical settings. This study explores a new index, the Longitudinal Asymmetry Index (ILong) which is derived using data from a low-cost depth camera (Kinect). ILong is based on depth images averaged over several gait cycles, rather than derived joint positions or angles. This study aims to evaluate (1) the validity of CRP computed with Kinect, (2) the validity and sensitivity of ILong for measuring gait asymmetry based solely on data provided by a depth camera, (3) the clinical applicability of a posteriorly mounted camera system to avoid occlusion caused by the standard front-fitted treadmill consoles and (4) the number of strides needed to reliably calculate ILong. The gait of 15 subjects was recorded concurrently with a marker-based system (MBS) and Kinect, and asymmetry was artificially reproduced by introducing a 5 cm sole attached to one foot. CRP computed with Kinect was not reliable. ILong detected this disturbed gait reliably and could be computed from a posteriorly placed Kinect without loss of validity. A minimum of five strides was needed to achieve a correlation coefficient of 0.9 between standard MBS and low-cost depth camera based ILong. ILong provides a clinically pragmatic method for measuring gait asymmetry, with application for improved patient care through enhanced disease, screening, diagnosis and monitoring.

Introduction

Lower-limb gait asymmetry (GA) is characteristic of a number of chronic neurological and musculoskeletal diseases [1], including cerebral palsy [2], Parkinson’s disease [3] and knee and hip osteoarthritis [4], [5]. Onset of GA can also follow acute events such as stroke [6] and amputation [7]. Thus, quantification of GA has become central to the clinical monitoring and treatment of many chronic and acute neurological and musculoskeletal diseases [8], [9].

Gait asymmetry is typically evaluated using kinematic data and spatio-temporal gait parameters (step/stride length and duration). Based on these global variables, indicators have been introduced to quantify asymmetry, including the Symmetry Ratio [6] and Symmetry Index [6]. These indicators provide global asymmetry information for a studied gait cycle, but cannot be used to analyse how asymmetry occurs at specific instances within the gait cycle.

To overcome this limitation, some authors have proposed analysis of the Constant Relative Phase (CRP) and its derivatives [10]. For a given joint, this indicator constructs a phase diagram based on angular value and angular velocity. Gait asymmetry can be assessed by comparing the CRP of the corresponding left and right joints. It has been used to measure asymmetry caused by weights placed on one foot [11] and pathologies such as stroke [12].

However, CRP is based on sagittal joint angles and their derivatives, which impose the use of accurate and noise-free motion capture systems, such as optoelectronic systems [11], [12]. Such systems are difficult to use in daily practice in small clinics as they require specific technical skills, calibration, and manual processing. In everyday practice, especially in small clinics, easy-to-use and low-cost systems are preferable, such as depth cameras developed for the gaming industries, which have been recently proposed for clinical gait analysis [13], [14]. However, the authors reported problems when computing temporal information due to inaccuracies and noise. These problems could impose important limitations when computing CRP, especially for the computation of reliable joint angular velocities but this has not yet been tested.

To tackle this potential problem, recent works proposed the direct use of depth images instead of skeleton data to perform clinical gait analysis. Accurate determination of global spatiotemporal data [15] and a continuous asymmetry index – the Longitudinal Asymmetry Index (ILong) [16] has been investigated using data obtained from low cost depth cameras such as the Kinect. ILong evaluates gait asymmetry at each instant of the gait cycle according to the instantaneous spatial distance between the right leg motion expressed during a stride starting with a right-foot strike, and the mirrored left leg motion expressed during a stride starting with a left-foot strike. To reduce noise, a representative gait cycle made of a sequence of averaged depth images (obtained with hundreds of gait cycles) was used to compute ILong. However, no benefit of ILong over CRP was investigated. Moreover, the authors validated this approach with a Kinect placed in front of a treadmill in order to capture and use numerous gait cycles. This was achieved using a specialised treadmill with a remotely placed console so as not to obscure the camera view. ILong computed from an anteriorly placed Kinect would not be applicable to standard treadmills available within a clinical setting, all of which have a front fitting console for safety reasons.

Therefore, this study aimed to evaluate (1) the validity and sensitivity of the CRP computed with Kinect, (2) validity and sensitivity of ILong for measuring gait asymmetry given the noisy data provided by Kinect, (3) the clinical applicability of ILong by placing the Kinect posteriorly rather than anteriorly to the treadmill, to reduce occlusion caused by front-fitted treadmill consoles, and (4) the number of strides needed to reliably calculate ILong.

The main hypotheses were:

-
H1: CRP computed with Kinect data is not reliable, supporting the idea of developing a noise-resistant continuous asymmetry index based on Kinect data,
-
H2: ILong computed with Kinect and Vicon data is valid and sensitive enough to detect asymmetrical gaits compared to natural ones,
-
H3: placing the Kinect behind the subject does not affect the validity of the ILong,
-
H4: the number of gait cycles averaged to compute ILong does not strongly affect the results, supporting the idea that a low number of cycles may be possible for some patients.

Section snippets

Methods

A protocol similar to that which has been described previously [16], [17] was carried out. Healthy subjects were asked to walk on a treadmill with unilateral imposed perturbation (a 5 cm sole placed below one of the feet) to induce gait asymmetry. CRP and ILong were then both computed with Kinect and Vicon data concurrently.

Results

Fig. 2 depicts CRP computed with Vicon and with Kinect data. As shown in Table 1A, CRP computed with Vicon data was able to detect asymmetry within the stride up to 95% of the time. However, CRP computed with Kinect data at the shank could only detect asymmetry for 73% (left deformation) and 52% (right deformation) within the stride. The results were similar when the CRP was computed at the thigh. Correlation between CRP computed with Vicon and Kinect data all along the cycle were 0.45 and 0.11

Discussion

The poor results obtained with the CRP computed with Kinect data supports hypothesis H1: CRP was not reliably computed from gait data and this could be explained by two reasons. Firstly, the derivation of noisy joint angles delivered by the Kinect induces noisy derivatives. Previous work demonstrated a correlation of 0.8 between joint angles delivered by the Kinect and an optoelectronic system [21]. This could partly explain the low correlation between CRP computed with Kinect and Vicon data.

Conclusion

This study shows that CRP computed with skeleton Kinect data failed to assess GA, whereas validity and sensitivity analysis demonstrated reliable assessment when using the new GA index (ILong) based on depth images. This provides the clinician with a longitudinal space-time GA index within the gait cycle for patients walking on a treadmill. We have shown that placing the depth camera in front of or behind the subject had no strong influence on the results and that even a small number of gait

Conflict of interest

The authors declare that they have no conflict of interest.

Acknowledgements

This work was funded by the Wellcome Trust and EPSRC grant number 088844/Z/09/Z and the Association nationale de la recherche et de la technologie (ANRT) (France) for the Conventions Industrielles de Formation par la REcherche (CIFRE) 146/2009 funding.

References (23)

M.P. Griffin et al.
Role of symmetry in gait performance of stroke subjects with hemiplegia
Gait Posture
(1995)
H. Böhm et al.
Gait asymmetries in children with cerebral palsy: do they deteriorate with running?
Gait Posture
(2012)
P. James et al.
A comparison of gait symmetry and hip movements in the assessment of patients with monoarticular hip arthritis
Clin. Biomech.
(1994)
K.K. Patterson et al.
Evaluation of gait symmetry after stroke: a comparison of current methods and recommendations for standardization
Gait Posture
(2010)
K.R. Kaufman et al.
Gait asymmetry of transfemoral amputees using mechanical and microprocessor-controlled prosthetic knees
Clin. Biomech.
(2012)
F. Liao et al.
Multi-resolution entropy analysis of gait symmetry in neurological degenerative diseases and amyotrophic lateral sclerosis
Med. Eng. Phys.
(2008)
P.F. Lamb et al.
On the use of continuous relative phase: review of current approaches and outline for a new standard
Clin. Biomech.
(2014)
S.A. Combs et al.
Bilateral coordination and gait symmetry after body-weight supported treadmill training for persons with chronic stroke
Clin. Biomech.
(2013)
R.A. Clark et al.
Concurrent validity of the Microsoft Kinect for assessment of spatiotemporal gait variables
J. Biomech.
(2013)
E. Auvinet et al.
Detection of gait cycles in treadmill walking using a Kinect
Gait Posture
(2015)

D. Gouwanda et al.

Identifying gait asymmetry using gyroscopes-a cross-correlation and normalized symmetry index approach

J. Biomech.

(2011)

Cited by (28)

Concurrent validity of a custom computer vision algorithm for measuring lumbar spine motion from RGB-D camera depth data
2021, Medical Engineering and Physics
Citation Excerpt :
RGB-D cameras are less expensive and require less time than traditional optoelectronic motion capture equipment for tracking human motion, and they can be easily installed within clinical settings. The performance and implementation of these cameras have been studied for the assessment of various experimental and clinical outcome measures relative to gold-standard motion tracking equipment, such as: 1) clinical measurements of motor function and postural control [12-14]; 2) spatiotemporal gait parameters during overground and treadmill walking [14-15]; and 3) spatial accuracy (i.e., location and orientation) of landmarks during various movement tasks [12,16-17]. The majority of studies that utilize RGB-D camera systems for tracking human motion achieve adequate results using the integrated skeletal model [12,15,18].
Using RGB-D cameras as an alternative motion capture device can be advantageous for biomechanical spine motion assessments of movement quality and dysfunction due to their lower cost and complexity. In this study, we evaluated RGB-D camera performance relative to gold-standard optoelectronic motion capture equipment. Twelve healthy young adults (6M, 6F) were recruited to perform repetitive spine flexion-extension, while wearing infrared reflective marker clusters placed over their T₁₀-T₁₂ spinous processes and sacrum, and motion capture data were recorded simultaneously by both systems. Custom computer vision algorithms were developed to extract spine angles from depth data. Root mean square error (RMSE) was calculated for continuous Euler angles, and intraclass correlation coefficients (ICC_2,1) were calculated between minimum and maximum angles and range of motion in all movement planes. RMSE was low (RMSE ≤ 2.05°) and reliability was good to excellent (0.849 ≤ ICC_2,1 ≤ 0.979) across all movement planes. In conclusion, the proposed algorithm for tracking 3D lumbar spine motion during a sagittal movement task from one RGB-D camera is reliable in comparison to gold-standard motion tracking equipment. Future research will investigate accuracy and validity in a wider variety of movements, and will also investigate the development of novel methods to measure spine motion without using infrared reflective markers.
Concurrent validity of evaluating knee kinematics using Kinect system during rehabilitation exercise
2021, Medicine in Novel Technology and Devices
Squats, ascending and descending stairs, and lunges are three functional movements popularly used in clinical rehabilitation. This study aimed to analyze the validity of the Kinect v2 for assessing the kinematics of three functional knee movements. An intelligent program was developed for real-time kinematic evaluation of the knee joint using the Kinect v2 system. Based on image recognition technology, images from the Kinect v2 depth camera were used to develop a kinematic evaluation program for knee joint movements. The kinematics of 30 healthy volunteers during stair ascent and descent, squatting, and lunging movements were collected synchronously with both VICON and Kinect systems. The peak knee angles in the sagittal and coronal planes were extracted and analyzed. A paired samples t-test and the relative coefficient within the groups were used for statistical analysis of the validity of the Kinect compared to the VICON. There was no significant difference (p > 0.05) in bilateral knee flexion angle when lunging. No significant difference (p > 0.05) was found in left knee flexion during squatting exercises. Although the Kinect generally underestimated knee flexion and valgus, it displayed excellent consistency and good to excellent agreement with the VICON motion analysis system. The results demonstrated that the Kinect based measurement tool for analyzing knee kinematics has the potential to be an effective clinical training device for maintaining knee alignment during exercise.
Effects of camera viewing angles on tracking kinematic gait patterns using Azure Kinect, Kinect v2 and Orbbec Astra Pro v2
2021, Gait and Posture
Citation Excerpt :
Gait patterns can be characterized using spatiotemporal (e.g., walking speed, step length, stance time) and kinematic measurements (e.g., joint angles) [4]. Underlying gait impairments can be identified by examining gait variability [5] and symmetry [6]. Three-dimension (3D) opto-electronic motion capture is considered the gold standard of gait analysis by the scientific community for reliable and accurate measurement of gait patterns [7].
Depth sensors could be a portable, affordable, marker-less alternative to three-dimension motion capture systems for gait analysis, but the effects of camera viewing angles on their joint angle tracking performance have not been fully investigated.
This study evaluated the accuracies of three depth sensors [Azure Kinect (AK); Kinect v2 (K2); Orbbec Astra (OA)] for tracking kinematic gait patterns during treadmill walking at five camera viewing angles (0°/22.5°/45°/67.5°/90°).
Ten healthy subjects performed fifteen treadmill walking trials (3 speeds × 5 viewing angles) using the three depth sensors to measure joint angles in sagittal hip, frontal hip, sagittal knee, and sagittal ankle. Ten walking steps were recorded and averaged for each walking trial. Range of motion in terms of maximum and minimum joint angles measured by the depth sensors were compared with the Vicon motion capture system as the gold standard. Depth sensors tracking accuracies were compared against the Vicon reference using root-mean-square error (RMSE) on the joint angle time series. Effects of different walking speeds, viewing angles, and depth sensors on the tracking accuracy were observed using three-way repeated-measure analysis of variance (ANOVA).
ANOVA results on RMSE showed significant interaction effects between viewing angles and depth sensors for sagittal hip [F(8,72) = 4.404, p = 0.005] and for sagittal knee [F(8,72)=13.211, p < 0.001] joint angles. AK had better tracking performance when subjects walked at non-frontal camera viewing angles (22.5°/45°/67.5°/90°); while K2 performed better at frontal viewing angle (0°). The superior tracking performance of AK compared with K2/OA might be attributed to the improved depth sensor resolution and body tracking algorithm.
Researchers should be cautious about camera viewing angle when using depth sensors for kinematic gait measurements. Our results demonstrated Azure Kinect had good tracking performance of sagittal hip and sagittal knee joint angles during treadmill walking tests at non-frontal camera viewing angles.
Fall risk assessment in the wild: A critical examination of wearable sensor use in free-living conditions
2021, Gait and Posture
Citation Excerpt :
A wide range of methods have been investigated to measure free-living mobility behaviour. Ambient sensors, such as radar [18], passive infrared [19], third-person video, and depth cameras [20–23] have been investigated as a means to extract gait parameters, detect falls, and track longitudinal changes in a person's mobility patterns. However, ambient sensors have limitations due to visual occlusions (e.g., furniture), inability to extract spatiotemporal data when full-body view is unavailable, and tracking the same person in spaces with multiple residents with similar body characteristics [24].
Despite advances in laboratory-based supervised fall risk assessment methods (FRAs), falls still remain a major public health problem. This can be due to the alteration of behavior in laboratory due to the awareness of being observed (i.e., Hawthorne effect), the multifactorial complex etiology of falls, and our limited understanding of human behaviour in natural environments, or in the’ wild’. To address these imitations, a growing body of literature has focused on free-living wearable-sensor-based FRAs. The objective of this narrative literature review is to discuss papers investigating natural data collected by wearable sensors for a duration of at least 24 h to identify fall-prone older adults.
Databases (Scopus, PubMed and Google Scholar) were searched for studies based on a rigorous search strategy.
Twenty-four journal papers were selected, in which inertial sensors were the only wearable system employed for FRA in the wild. Gait was the most-investigated activity; but sitting, standing, lying, transitions and gait events, such as turns and missteps, were also explored. A multitude of free-living fall predictors (FLFPs), e.g., the quantity of daily steps, were extracted from activity bouts and events. FLFPs were further categorized into discrete domains (e.g., pace, complexity) defined by conceptual or data-driven models. Heterogeneity was found within the reviewed studies, which includes variance in: terminology (e.g., quantity vs macro), hyperparameters to define/estimate FLFPs, models and domains, and data processing approaches (e.g., the cut-off thresholds to define an ambulatory bout). These inconsistencies led to different results for similar FLFPs, limiting the ability to interpret and compare the evidence.
Free-living FRA is a promising avenue for fall prevention. Achieving a harmonized model is necessary to systematically address the inconsistencies in the field and identify FLFPs with the highest predictive values for falls to eventually address intervention programs and fall prevention.
Simplified digital balance assessment in typically developing school children
2021, Gait and Posture
Citation Excerpt :
Against VICON™, the Kinect™ demonstrated > 90 % spatial accuracy for movements in a frontal plane, and 83 % for depth [25]. Multiple studies demonstrated the validity of markerless 3DMA to assess postural stability [8–10] and gait abnormalities [26]. While the accuracy in tracking smaller joints such as the ankle or wrist is reported lower than for central joints and trunk [27], the reliability in detecting and tracking kinematics even in small children was found to be moderate to good [28].
Postural balance can be considered a conjoined parameter of gross motor performance. It is acquired in early childhood and honed until adolescence, but may also be influenced by various conditions. A simplified clinical assessment of balance and posture could be helpful in monitoring motor development or therapy particularly in pediatric patients. While analogue scales are considered unprecise and lab-based force-plate posturography lacks accessibility, we propose a novel kinematic balance assessment based on markerless 3D sensor technology.
Can balance and posture be assessed by tracking kinematic data using a single 3D motion tracking camera and are the results representative of normal motor development in a healthy pediatric cohort?
A proprietary algorithm was developed and tested that uses skeletal data from the Microsoft Kinect™ V2 3D motion capture camera to calculate and track the center of mass in real time during a set of balance tasks. The algorithm tracks the distance of the COM traveled over time to calculate a balance score (COM speed). For this study, 432 school children aged 4−18 years performed 5 balance tasks and the resulting balance scores were analyzed and correlated with demographic data.
Preliminary experiments demonstrated that the system was able to reliably detect differences in COM speed during different balance tasks. The method showed moderate correlation with age and sex. Athletic activity positively correlated with balance skill in the age group < 8 years, but not in older children. Body mass appeared not to be correlated with balance ability.
This study demonstrates that markerless 3D motion analysis can be used for the clinical assessment of coordination and balance and could potentially be used to monitor gross motor performance at the point-of-care.
SpiderNet: A spiderweb graph neural network for multi-view gait recognition
2020, Knowledge-Based Systems
Citation Excerpt :
In this way, it can only be explored from the multi-modal identification of data. At present, some visual sensors, such as depth camera, infrared camera, and binocular camera [2–5], acquire different types of gait images, which are further analyzed and recognized using fusion algorithms. But it can only express features from a single angle, ignoring the gait view observed from other angles.
Human gait is a proven biometric trait with applications in security for authentication and disease diagnosis. However, it is one-sided to express and interpret gait data from a single point of view, which cannot reflect multi-dimensional characteristics of gait changes. Moreover, if the gait pattern observed from other views has pathological or abnormal behavior, or has micro movement, it is not easy to be detected and thus affects the recognition rate of gait. In addition, the multi-view fusion of gait knowledge can be challenging due to the close correlation between various visual angles. Owing to the above facts, we propose a spiderweb graph neural network (SpiderNet) to solve the multi-view gait recognition problem, which connects the gait data of single view with that of other views concurrently and constructs an active graph convolutional neural network. The gait trajectory of each view is analyzed by the combination of a memory module and a capsule module, which accomplishes the multi-view feature fusion, as well as the spatio-temporal feature extraction of single view. The experimental results show that the SpiderNet is superior to fifteen state-of-the-art methods, such as random forest, long-short term memory and convolutional neural network, and achieves 98.54%, 98.77%, and 96.91% of the results on three challenging gait datasets: SDUgait, CASIA-B, and OU-MVLP.

View all citing articles on Scopus

View full text

Full length articleValidity and sensitivity of the longitudinal asymmetry index to detect gait asymmetry using Microsoft Kinect data

Highlights

Abstract

Introduction

Section snippets

Methods

Results

Discussion

Conclusion

Conflict of interest

Acknowledgements

Gait Posture

Gait Posture

Clin. Biomech.

Gait Posture

Clin. Biomech.

Med. Eng. Phys.

Clin. Biomech.

Clin. Biomech.

J. Biomech.

Gait Posture

J. Biomech.

Full length article
Validity and sensitivity of the longitudinal asymmetry index to detect gait asymmetry using Microsoft Kinect data