Inertial Sensor-Based Gait Recognition: A Review

A ROBUST AUTOMATIC GAIT MONITORING APPROACH USING A SINGLE IMU FOR HOME-BASED APPLICATIONS

A new approach of human activity monitoring with a single Inertial Measurement Unit (IMU) capable of gait recognition and assessment is proposed for home-based applications. The method estimates the foot motion using a single IMU, then automatically segments the motion into steps, and extracts multiple kinematics templates. It classifies each segment by extracting Mahalanobis distance-based features from multiple sections of the motion templates and then training a Support Vector Machine. The proposed wearable system can distinguish between nine classes of activities with a classification accuracy of 99.6%. It can also discriminate between normal and abnormal gait patterns with an accuracy of 98.7%. In addition to a high recognition rate, the proposed approach provides a Gait Similarity Score (GSS) of the performed gait to its desired/normal pattern. The experimental results indicate the capability of GSS measure for assessing the quality of motion in “pre-”, “initial”, “mid” and “terminal” stages of swing phase.

Inertial measurement systems for segments and joints kinematics assessment: towards an understanding of the variations in sensors accuracy

BACKGROUND

Joints kinematics assessment based on inertial measurement systems, which include attitude and heading reference system (AHRS), are quickly gaining in popularity for research and clinical applications. The variety of the tasks and contexts they are used in require a deep understanding of the AHRS accuracy for optimal data interpretation. However, published accuracy studies on AHRS are mostly limited to a single task measured on a limited number of segments and participants. This study assessed AHRS sensors kinematics accuracy at multiple segments and joints through a variety of tasks not only to characterize the system's accuracy in these specific conditions, but also to extrapolate the accuracy results to a broader range of conditions using the characteristics of the movements (i.e. velocity and type of motion). Twenty asymptomatic adults ([Formula: see text] = 49.9) performed multiple 5 m timed up and go. Participants' head, upper trunk, pelvis, thigh, shank and foot were simultaneously tracked using AHRS and an optical motion capture system (gold standard). Each trial was segmented into basic tasks (sit-to-stand, walk, turn).

RESULTS

At segment level, results revealed a mean root-mean-squared-difference [Formula: see text] varying between 1.1° and 5.5° according to the segment tracked and the task performed, with a good to excellent agreement between the systems. Relative sensor kinematics accuracy (i.e. joint) varied between 1.6° and 13.6° over the same tasks. On a global scheme, analysis of the effect of velocity on sensor kinematics accuracy showed that AHRS are better adapted to motions performed between 50°/s and 75°/s (roughly thigh and shank while walking).

CONCLUSION

Results confirmed that pairing of modules to obtain joint kinematics affects the accuracy compared to segment kinematics. Overall, AHRS are a suitable solution for clinical evaluation of biomechanics under the multi-segment tasks performed although the variation in accuracy should be taken into consideration when judging the clinical meaningfulness of the observed changes.

Evaluating physical function and activity in the elderly patient using wearable motion sensors

Wearable sensors, in particular inertial measurement units (IMUs) allow the objective, valid, discriminative and responsive assessment of physical function during functional tests such as gait, stair climbing or sit-to-stand.

Applied to various body segments, precise capture of time-to-task achievement, spatiotemporal gait and kinematic parameters of demanding tests or specific to an affected limb are the most used measures.

In activity monitoring (AM), accelerometry has mainly been used to derive energy expenditure or general health related parameters such as total step counts.

In orthopaedics and the elderly, counting specific events such as stairs or high intensity activities were clinimetrically most powerful; as were qualitative parameters at the ‘micro-level’ of activity such as step frequency or sit-stand duration.

Low cost and ease of use allow routine clinical application but with many options for sensors, algorithms, test and parameter definitions, choice and comparability remain difficult, calling for consensus or standardisation.

Cite this article: Grimm B, Bolink S. Evaluating physical function and activity in the elderly patient using wearable motion sensors. EFORT Open Rev 2016;1:112–120. DOI: 10.1302/2058-5241.1.160022.

Wearable Device-Based Gait Recognition Using Angle Embedded Gait Dynamic Images and a Convolutional Neural Network

The widespread installation of inertial sensors in smartphones and other wearable devices provides a valuable opportunity to identify people by analyzing their gait patterns, for either cooperative or non-cooperative circumstances. However, it is still a challenging task to reliably extract discriminative features for gait recognition with noisy and complex data sequences collected from casually worn wearable devices like smartphones. To cope with this problem, we propose a novel image-based gait recognition approach using the Convolutional Neural Network (CNN) without the need to manually extract discriminative features. The CNN’s input image, which is encoded straightforwardly from the inertial sensor data sequences, is called Angle Embedded Gait Dynamic Image (AE-GDI). AE-GDI is a new two-dimensional representation of gait dynamics, which is invariant to rotation and translation. The performance of the proposed approach in gait authentication and gait labeling is evaluated using two datasets: (1) the McGill University dataset, which is collected under realistic conditions; and (2) the Osaka University dataset with the largest number of subjects. Experimental results show that the proposed approach achieves competitive recognition accuracy over existing approaches and provides an effective parametric solution for identification among a large number of subjects by gait patterns.

Validation of Energy Expenditure Prediction Models Using Real-Time Shoe-Based Motion Detectors

OBJECTIVE

This study aimed to verify and compare the accuracy of energy expenditure (EE) prediction models using shoe-based motion detectors with embedded accelerometers.

METHODS

Three physical activity (PA) datasets (unclassified, recognition, and intensity segmentation) were used to develop three prediction models. A multiple classification flow and these models were used to estimate EE. The "unclassified" dataset was defined as the data without PA recognition, the "recognition" as the data classified with PA recognition, and the "intensity segmentation" as the data with intensity segmentation. The three datasets contained accelerometer signals (quantified as signal magnitude area (SMA)) and net heart rate (HR_net). The accuracy of these models was assessed according to the deviation between physically measured EE and model-estimated EE.

RESULTS

The variance between physically measured EE and model-estimated EE expressed by simple linear regressions was increased by 63% and 13% using SMA and HR_net, respectively. The accuracy of the EE predicted from accelerometer signals is influenced by the different activities that exhibit different count-EE relationships within the same prediction model.

CONCLUSION

The recognition model provides a better estimation and lower variability of EE compared with the unclassified and intensity segmentation models.

SIGNIFICANCE

The proposed shoe-based motion detectors can improve the accuracy of EE estimation and has great potential to be used to manage everyday exercise in real time.

Cross View Gait Recognition Using Joint-Direct Linear Discriminant Analysis

This paper proposes a view-invariant gait recognition framework that employs a unique view invariant model that profits from the dimensionality reduction provided by Direct Linear Discriminant Analysis (DLDA). The framework, which employs gait energy images (GEIs), creates a single joint model that accurately classifies GEIs captured at different angles. Moreover, the proposed framework also helps to reduce the under-sampling problem (USP) that usually appears when the number of training samples is much smaller than the dimension of the feature space. Evaluation experiments compare the proposed framework's computational complexity and recognition accuracy against those of other view-invariant methods. Results show improvements in both computational complexity and recognition accuracy.

A novel accelerometry-based algorithm for the detection of step durations over short episodes of gait in healthy elderly

Background

The assessment of short episodes of gait is clinically relevant and easily implemented, especially given limited space and time requirements. BFS (body-fixed-sensors) are small, lightweight and easy to wear sensors, which allow the assessment of gait at relative low cost and with low interference. Thus, the assessment with BFS of short episodes of gait, extracted from dailylife physical activity or measured in a standardised and supervised setting, may add value in the study of gait quality of the elderly. The aim of this study was to evaluate the accuracy of a novel algorithm based on acceleration signals recorded at different human locations (lower back and heels) for the detection of step durations over short episodes of gait in healthy elderly subjects.

Methods

Twenty healthy elderly subjects (73.7 ± 7.9 years old) walked twice a distance of 5 m, wearing a BFS on the lower back, and on the outside of each heel. Moreover, an optoelectronic three-dimensional (3D) motion tracking system was used to detect step durations. A novel algorithm is presented for the detection of step durations from low-back and heel acceleration signals separately. The accuracy of the algorithm was assessed by comparing absolute differences in step duration between the three methods: step detection from the optoelectronic 3D motion tracking system, step detection from the application of the novel algorithm to low-back accelerations, and step detection from the application of the novel algorithm to heel accelerations.

Results

The proposed algorithm successfully detected all the steps, without false positives and without false negatives. Absolute average differences in step duration within trials and across subjects were calculated for each comparison, between low-back accelerations and the optoelectronic system were on average 22.4 ± 7.6 ms (4.0 ± 1.3 % of average step duration), between heel accelerations and the optoelectronic system were on average 20.7 ± 11.8 ms (3.7 ± 1.9 %), and between low-back accelerations and heel accelerations were on average 27.8 ± 15.1 ms (4.9 ± 2.5 % of average step duration).

Conclusions

This study showed that the presented novel algorithm detects step durations over short episodes of gait in healthy elderly subjects with acceptable accuracy from low-back and heel accelerations, which provides opportunities to extract a range of gait parameters from short episodes of gait.