Robotics‐driven gait analysis: Assessing Azure Kinect's performance in in‐lab versus in‐corridor environments
Diego Guffanti,Alberto Brunete,Miguel Hernando,David Álvarez,Ernesto Gambao,William Chamorro,Diego Fernández‐Vázquez,Víctor Navarro‐López,María Carratalá‐Tejada,Juan Carlos Miangolarra‐Page
DOI: https://doi.org/10.1002/rob.22313
IF: 6.385
2024-03-15
Journal of Field Robotics
Abstract:Gait analysis offers vital insights into human movement, aiding in the diagnosis, treatment, and rehabilitation of various conditions. Analyzing gait in corridors, rather than in lab, provides unique advantages for a more comprehensive understanding of human locomotion. However, limited dedicated technologies constrain gait data analysis in this context. In this study, a markerless gait analysis system using an Azure Kinect sensor mounted on a mobile robot is proposed and validated as a potential solution for gait analysis in corridors. Ten healthy participants (4 males and 6 females) underwent two tests. The first test (5 trials per participant) took place in the laboratory. Here, Azure Kinect performance was validated against a Vicon system, assessing eight gait signals and 22 gait parameters. The second test (2 trials per participant) was performed in the corridors over a 32‐m walking distance to compare this gait pattern with the one developed within the laboratory. The intrasession Intraclass Correlation Coefficient (ICC) reliability for in‐lab experiments was assessed by calculating the ICC between gait cycles captured in each session per participant. Notably, knee flexion/extension (ICC‐0.95), hip flexion/extension (ICC‐0.96), pelvis rotation (ICC‐0.88), and interankle distance (ICC‐0.98) demonstrated excellent reliability with high confidence. Similarly, hip adduction/abduction showed good reliability (ICC‐0.79), while trunk rotation exhibited moderate reliability (ICC‐0.72). In contrast, both trunk tilt (ICC‐0.24) and pelvis tilt (ICC‐0.41) consistently displayed lower reliability. This was observed for both the Vicon and the Azure systems, highlighting the intricate nature of capturing precise data for these specific signals in both systems. Validity outcomes indicated comparable error rates to literature standards (12.68 ° knee flexion/extension, 5.54 ° hip flexion/extension, and 3.45 ° hip adduction/abduction), with 11 parameters having no significant differences from Vicon. Comparison of in‐lab and in‐corridor experiments show that individuals exhibit significantly longer stride time (1.10 s vs. 1.05 s), lower pelvis tilt (6.83 ° vs. 9.39 ° ), and lower minimum pelvis rotation (− 5.82 ° vs. − 14.61 ° ) when walking in the laboratory. This study demonstrates promising outcomes in outdoor gait analysis with a robot‐mounted camera, revealing significant distinctions from controlled laboratory evaluations
robotics