A Low-Cost Computer Vision Approach for Objectively Evaluating Triple Jump Technique: Proof of Concept

Abstract

The triple jump is a track and field event comprising an approach run followed by three phases – hop, step, and jump – culminating in a sand pit landing. Typically, triple jump technique is evaluated subjectively by coaches, with quantitative assessments often relying on high-cost multi-camera systems. To address this limitation, a low-cost framework was developed for objectively analyzing triple jump technique in real-world settings using a single iPhone 12 camera and the open-source human pose estimation software, Google MediaPipe pose (GMP pose). Data were collected from 30 participants (15 male, 15 female) performing six triple jumps each across four different track and field facilities, resulting in 154 usable videos for analysis. Participants were categorized by skill level based on their personal best triple jump distance. Central to the framework is the use of homography transformations, which enabled the conversion from GMP pose data to real-world coordinates to calculate triple jump performance metrics, including horizontal and vertical centre of mass velocities, last stride velocity, effective distance, phase distances and phase ratios. Additional performance metrics, including ground contact times, flight times, and knee angles, were derived from the video recordings and GMP pose data. Validation of the pose estimations and homography transformations was done primarily through subjective visual assessments using a custom Pythontool, with reprojection error serving as an additional quantitative measure of relative homography performance. The performance metric results showed that more skilled participants exhibited superior step phases compared to their less skilled counterparts and demonstrated performance trends similar to elite jumpers from World Athletics reports, particularly in the reduction of horizontal and vertical velocity across jump phases, albeit at lower velocities. Moreover, participants’ phase ratios were consistent with elite-level jumpers. Comparisons with World Athletics data provided a reference to validate the triple jump performance metrics, showing promising alignment. This study demonstrates the potential of an affordable, single-camera markerless motion capture system to analyze triple jump technique in real-world settings. Although the system shows promise, much of the process remains manual, limiting its scalability for broader use. Despite its current limitations, the proposed framework offers a novel perspective on the feasibility of affordable, single camera markerless motion capture systems in real-world environments, providing a foundation for further development.