Machine-learning-based children's pathological gait classification with low-cost gait-recognition system - PubMed Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2021 Jun 22;20(1):62.
doi: 10.1186/s12938-021-00898-0.

Machine-learning-based children's pathological gait classification with low-cost gait-recognition system

Linghui Xu  1   2 , Jiansong Chen  3 , Fei Wang  4 , Yuting Chen  5 , Wei Yang  6   7 , Canjun Yang  1   2
Affiliations

Machine-learning-based children's pathological gait classification with low-cost gait-recognition system

Linghui Xu et al. Biomed Eng Online. .

Abstract

Background: Pathological gaits of children may lead to terrible diseases, such as osteoarthritis or scoliosis. By monitoring the gait pattern of a child, proper therapeutic measures can be recommended to avoid the terrible consequence. However, low-cost systems for pathological gait recognition of children automatically have not been on market yet. Our goal was to design a low-cost gait-recognition system for children with only pressure information.

Methods: In this study, we design a pathological gait-recognition system (PGRS) with an 8 × 8 pressure-sensor array. An intelligent gait-recognition method (IGRM) based on machine learning and pure plantar pressure information is also proposed in static and dynamic sections to realize high accuracy and good real-time performance. To verifying the recognition effect, a total of 17 children were recruited in the experiments wearing PGRS to recognize three pathological gaits (toe-in, toe-out, and flat) and normal gait. Children are asked to walk naturally on level ground in the dynamic section or stand naturally and comfortably in the static section. The evaluation of the performance of recognition results included stratified tenfold cross-validation with recall, precision, and a time cost as metrics.

Results: The experimental results show that all of the IGRMs have been identified with a practically applicable degree of average accuracy either in the dynamic or static section. Experimental results indicate that the IGRM has 92.41% and 97.79% intra-subject recognition accuracy, and 85.78% and 78.81% inter-subject recognition accuracy, respectively, in the static and dynamic sections. And we find methods in the static section have less recognition accuracy due to the unnatural gesture of children when standing.

Conclusions: In this study, a low-cost PGRS has been verified and realize feasibility, highly average precision, and good real-time performance of gait recognition. The experimental results reveal the potential for the computer supervision of non-pathological and pathological gaits in the plantar-pressure patterns of children and for providing feedback in the application of gait-abnormality rectification.

Keywords: Feature extraction; Gait classification; Pathological gait recognition; Pressure-sensor array.

PubMed Disclaimer

Conflict of interest statement

The authors declare that they have no competing interests.

Figures

Fig. 1
Fig. 1
a Diagram of the gait-recognition algorithm. b Gait pattern for classification—toe-in, toe-out, flat and normal
Fig. 2
Fig. 2
The procedure of gait feature extraction from gait cycle. a 30 sensor blocks with the red mark are selected for feature extraction. b Sliding window method (Hanning window with 512-sample intervals width) is applied to all n sensor blocks and transformed to frequency domain later. c Dividing five frequency bands from FFT frequency spectrum, 0 (exclude)–2 Hz, 2 (exclude)–4 Hz, 4 (exclude)–6 Hz, 6 (exclude)–8 Hz, and 8 (exclude)–10 Hz, and summing all the frequent value in each frequency band to generate a five-elements vector for each sensor. d Joining all the five-element vector from each sensor together and normalizing to a 150-element unit vector
Fig. 3
Fig. 3
a LDA and PCA algorithm performance in classification in the specific situation. In this situation, LDA’s performance is better than PCA’s. b SVM classification. The main idea of the SVM is projecting data points into a higher dimensional space, specified by a kernel function, and computing a maximum-margin hyperplane decision surface that separates the two classes
Fig. 4
Fig. 4
Pathological gait-recognition system. 8 × 8 sensor array is under the slipper and controller attached to the child’s leg with hook-and-loop fasteners. The control circuit board contains a signal-collecting circuit, a low-energy Bluetooth device (HC-42 with Bluetooth 5.0, HuiCheng Information Technology Co., Ltd., China), STM32F103 controller, and two 4.2-V Li-ion batteries. The signal-collecting circuit operates with 5 V of power generated by an LM7805 unit (KIA7805AP, three-terminal positive voltage regulator of 5 V, KEC, China) and the STM32F103 circuitry operates with 3.3 V generated by an AMS117 unit (low-dropout-voltage regulator with fixed 3.3 V, Advanced Monolithic Systems, Inc.)
Fig. 5
Fig. 5
a The three-layer structure of the piezoresistive sensor. b Piezoresistive sensor pressure–resistor characteristic curve. Horizontal axis represents the pressure loaded on the sensor and vertical axis is the resistor of sensor. c Sensor calibration. Same color points represent a certain sensor block output during calibration and regression straight line with the same color is its result. A total of 64 sensor block calibration results are shown
Fig. 6
Fig. 6
Piezoresistive sensor array scanner electronic schematic. a Schematic before applying elimination crosstalk method. A microcontroller is used to select the sampling row and column channel by controlling quad bilateral switch CD4066 chip, while other unselected channels are remaining high resistance. Crosstalk output is found between sensor blocks as path 1 and path 2 show. b Schematic after applying elimination crosstalk method. When a certain row and column channels are selected (row channel 1 and column channel 1 is selected here), we pull down other row channels to ground and pull up other column channels to Vref, such that path 2 and path 3 will be cut off and path 1 will remain. Crosstalk output between sensor blocks can be eliminated. c The load force on the sensor block at (2,3) has less influence on other sensors' value output after sensor array applying elimination crosstalk. The dotted line represents data after sensor array applying elimination crosstalk and the solid line represents data before sensor array applying elimination crosstalk
Fig. 7
Fig. 7
GUI background flowchart. GUI/PC can receive commands from people in the GUI and use BLE to communicate with the controller system
Fig. 8
Fig. 8
Experiment in the dynamic section: a snapshots of subjects; b experimental procedure of the dynamic section
Fig. 9
Fig. 9
Plantar-pressure data acquisition results. a Curve of plantar pressure with time during level-ground walking in the dynamic-section experiment. Subfigure on the left top shows the placement of the sensor corresponding to data with the same color. b Typical toe-in, toe-out, and normal foot-pressure distribution in the static section experiment. The plantar pressure value is expressed in different grey scales. Pure black means zero pressure, pure white means the largest pressure
Fig. 10
Fig. 10
Accuracy and time cost results of IGRM in the dynamic and static experiments. The x-axis label means reduction algorithm + classification algorithm. For example, LDA + SVMlin means that IGRM’s feature reduction algorithm is LDA, and SVM with linear kernel is its classification algorithm. Here, PCA components are 7 (capture 90% of the variance) in the dynamic section and 12 (capture 90% of the variance) in the static section. And the dimension of LDA is 3 in both dynamic and static section
Fig. 11
Fig. 11
Average confusion matrix of algorithm results in tenfold cross-validation of a dynamic and b static sections. The labels on the column are predicted labels and those on the row are actual labels; the samples number at the row label i and the column label j means the average number of times instances of class i are classified as class j in the tenfold procedure
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

References

    1. Figueiredo J, Santos CP, Moreno JC. Automatic recognition of gait patterns in human motor disorders using machine learning: A review. Med Eng Phys. 2018;53:1–12. doi: 10.1016/j.medengphy.2017.12.006. - DOI - PubMed
    1. Titianova EB, Mateev PS, Tarkka IM. Footprint analysis of gait using a pressure sensor system. J Electromyogr Kinesiol. 2004;14:275–281. doi: 10.1016/S1050-6411(03)00077-4. - DOI - PubMed
    1. Salarian A, Russmann H, Vingerhoets FJG, Dehollain C, Blanc Y, Burkhard PR, et al. Gait Assessment in Parkinson’s disease: toward an ambulatory system for long-term monitoring. IEEE Trans Biomed Eng. 2004;51:1434–1443. doi: 10.1109/TBME.2004.827933. - DOI - PubMed
    1. Chau T. A review of analytical techniques for gait data Part 1: Fuzzy, statistical and fractal methods. Gait Posture. 2001;13:49–66. doi: 10.1016/s0966-6362(00)00094-1. - DOI - PubMed
    1. Dolatabadi E, Taati B, Mihailidis A. An automated classification of pathological gait using unobtrusive sensing technology. IEEE Trans Neural Syst Rehabil Eng. 2017;25:2336–46. doi: 10.1109/tnsre.2017.2736939. - DOI - PubMed

LinkOut - more resources