Abstract

Introduction: Running is a common form of exercise that comes with many health benefits; however, many people stop exercising because of the high injury risk. Running injury is connected with biomechanical risk factors. Yet, these risk factors often show inconsistent and conflicting results across studies, largely due to small sample sizes limiting statistical power. Low subject numbers may be partly due to the time requirements of traditional motion capture. Monocular pose estimation offers a promising alternative by using machine learning to track body landmarks from a single camera. This study evaluated the feasibility of remotely collecting running biomechanics data using monocular pose estimation while exploring associations between biomechanics with injury and performance. Methods: Participants were recruited via social media and flyers at running stores. They completed a survey on injury history and performance, then ran on a treadmill at 7.5 mph while recording themselves in high-speed video for 30 seconds. Mediapipe Pose was used to track body landmarks. Feasibility was determined for recruitment adherence, and data collection. Logistic regressions were run between four injury groups (Bone injury, Lower leg injuries, Knee injuries, and Upper leg injuries) and running mechanics. Linear regressions were also run between the biomechanical variables and race times. Results: A total of 310 people participated 69.17 ± 0.86 kg; 1.75 ± 0.09m, 13.25 ±10.36 years of running, 33.36 ± 20.55 miles per week). Of those we received videos from 69 of them and were only able to include 45 of them in the final analysis. Significant predictors of prior injury were found in all but one injury category and in three of the four race distances. Discussion: Monocular pose estimation can be used for the remote collection of running mechanics data but poses several key concerns. The number of subjects in the final analysis was low due to upload failures and future research needs to determine better ways to receive large video files for analysis. Despite these issues, this study found significant predictors of injury and performance consistent with previous literature. There were also significant predictors identified that have not been studied extensively that may warrant further research.

Degree

MS

College and Department

Life Sciences; Exercise Sciences

Rights

https://lib.byu.edu/about/copyright/

Date Submitted

2025-08-12

Document Type

Thesis

Keywords

running, markerless motion capture, injury, performance

Language

english

Download

Included in

Life Sciences Commons

Share

COinS