A computer vision-based mobile tool for assessing human posture: A validation study

Non-invasive methods for postural assessment are tools used for tracking and monitoring the progression of postural deviations. Different computer-based methods have been used to assess human posture, including mobile applications based on images and sensors. However, such solutions still require manual identification of anatomical points. This study aims to present and validate the NLMeasurer, a mobile application for postural assessment. This application takes advantage of the PoseNet, a solution based on computer vision and machine learning used to estimate human pose and identify anatomical points. From the identified points, NLMeasurer calculates postural measures.

Twenty participants were photographed in front view while using surface markers over anatomical landmarks. Then, the surface markers were removed, and new photos were taken. The photos were analyzed by two examiners, and six postural measurements were computed with NLMeasurer and a validated biophotogrammetry software. One-sample t-test and Bland Altman procedure were used to assess agreement between the methods, and Intraclass Correlation Coefficient (ICC) was used to assess inter- and intra-rater reliability.

Postural measurements calculated using the NLMeasurer were in agreement with the biophotogrammetry software. Furthermore, there was good inter- and intra-rater reliability for most photos without surface markers.

NLMeasurer demonstrated to be a valid tool method to assess postural measurements in the frontal view. The use of surface markers on specific anatomical landmarks (i.e., ears, iliac spines and ankles) can facilitate the digital identification of these landmarks and improve the reliability of the postural measurements performed with NLMeasurer.