Pattern recognition of human movements using features extracted by triangulation method — a comparison with features of time and frequency domains

• Published in: Research on biomedical engineering Vol. 38; no. 2; pp. 539 - 557
• Main Authors: Giacomossi, Luiz Carlos, Pichorim, Sérgio Francisco
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.06.2022
• Subjects: Biomaterials; Biomedical Engineering and Bioengineering; Biomedical Engineering/Biotechnology; Engineering; Original Article; Frequency domain features; Pattern recognition; Time domain features; Human movement acceleration; Smartphone; Triangulation technique
• ISSN: 2446-4740; 2446-4740
• DOI: 10.1007/s42600-022-00204-4

Purpose In this article, a method for extracting features is presented, where an original signal is reduced and converted into a sequence of triangles based on peaks and valleys points. The patterns are formed by features obtained from the calculation of straight lines, angles, areas, slope, and total of triangles extracted from these sequences. This triangulation technique has the advantage of creating features that are simultaneously parameters in the domains of time and frequency. In calculating Euclidean distances or sides of triangles, both values of amplitude and time coordinates are employed. Methods Two algorithms were elaborated based on the triangulation techniques and compared with algorithms composed of classic features in the domains of time and frequency. Patterns were calculated for different sizes of input data window and quantity of features in three experiments developed to evaluate the triangulation method. The classification indexes of some human movement accelerations (walking, running, simulated tremors, clapping, and waving goodbye, acquired by smartphone) were used to compare these algorithms. Results Comparing the results obtained from the individual hit rate indexes per movements (studied categories), the triangulation method obtained the highest indexes for the walking and running movements. The average hit rates of 99.64% and 99.60% were achieved with the triangulation method and with the classic features, respectively, for windows of 200 and 500 points, with 36 features. Conclusion This research presented promising results in the classification of human movements applying the triangulation method, in view of the high average and individual hit rates.