Development of a system for analysis and recognition of human gait patterns with the application of clustering algorithms
Year: 2024
Type: dissertation
Abstract: This study presents the development of GAITWell, a human gait analysis system based on binary sensors and clustering algorithms. The system was designed to identify and classify gait patterns based on spatiotemporal variables, providing automatic results. GAITWell aims to overcome the limitations of traditional technologies, such as high costs and low accessibility, offering a viable alternative for middle- and low-income countries. Gait analysis systems provide valuable information on locomotor patterns, essential for diagnosing and treating disorders, which justifies the development of GAITWell. The objective of this study was to develop and implement a gait analysis system using binary sensors (logical levels 0 and 1) and clustering algorithms to identify locomotor patterns. GAITWell integrates data acquisition, control, footprint segmentation, database storage, and a graphical interface for clinical application. The software was developed in C# (.NET Framework) with SQLite storage. The hardware consists of sensors arranged on 44 × 37 cm plates (11 × 7 matrix with 4 cm spacing), designed to capture gait events and integrated into the system for analysis. Test-retest reliability was assessed using the intraclass correlation coefficient (ICC), standard error of measurement (SEM), and Bland-Altman plots. Concurrent validity was established by comparing GAITWell to Qualisys Pro-Reflex. The results indicated high accuracy in step identification, with moderate to high correlations for most gait variables. However, step length showed a lower correlation (r = 0.360, p < 0.05). The ICC2,1 demonstrated moderate to good agreement for most variables, except for double support time, cadence, and base of support, which exhibited lower agreement, indicating the need for improvements. The study involved 38 healthy adults (mean age: 33.2 years, SD 13.0; mean BMI: 26.6 kg/m², SD 5.4). Despite the high accuracy in footprint identification, some variables showed lower agreement, suggesting the need for improvement in gait event capture. The analysis revealed that sensor resolution directly impacts the accuracy of spatiotemporal variable determination. To mitigate this limitation, the impact of increasing sensor resolution to 1.2 cm × 1.2 cm was analyzed, which would reduce measurement error from 2 cm to 0.6 cm, representing a theoretical reduction of 70%. GAITWell has great potential for clinical applications and research, especially in resource-limited environments such as rehabilitation clinics and elderly care institutions. Its portability, modularity, and efficiency in data collection make it a fast and accessible alternative compared to traditional methods. To enhance its accuracy and applicability, the use of higher-resolution sensors and further validations with different population profiles are recommended. Additionally, its scalability and integration with other systems expand its potential for application in various clinical and research contexts. With these improvements, GAITWell has the potential to reduce costs and promote greater accessibility in the diagnosis and monitoring of gait disorders.
Author: Wellingtânia Domingos Dias
Institution: Universidade Federal de Minas Gerais
Language: Portuguese
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Keywords: gait analysis, gait pattern recognition, gait event detection, clustering algorithms, spatiotemporal gait parameters, test-retest reliability, concurrent validity, sensor resolution, locomotor disorders, resource-limited settings, rehabilitation centers, elderly care, step length, double support time, cadence
Open Access status: green
OpenAlex ID: W7119380046
