Uplift's Site

(Study completed October 1, 2024)

‍
Ricky Pimentel
Uplift Labs
1 October 2024

Background

Our goal is to test the accuracy of Uplift’s single camera (SC) algorithm compared to a force platform system (VALD Force Decks) for calculating countermovement jump metrics. We analyzed the event detection algorithms and jump metrics driven by Uplift’s movement analysis models.

Methods

Metrics and Events to Analyze We compare outcomes for:

‍Jump Height

‍Our primary measure is jump height, as that is the main outcome for countermovement jumps.

‍Flight Time

‍Jump height is derived from flight time, we also benchmark this outcome to ensure similarity across measures.

‍Modified Reactive Strength Index (RSI-mod)

‍RSI-mod is the ratio of jump height divided by the time from jump initiation until takeoff. It is a measure of one’s ability to generate power quickly and efficiently.

Protocol

10 elite level baseball players completed the jumps while simultaneously being recorded with high speed video and the VALD Force Decks system. Athletes completed six jumps in succession with a 15-20 second break between jumps. Athletes completed the first 3 jumps with their hands on hips and the next 3 jumps with normal arm swing. All successful jumps required jumping from and landing back on the force plates.

Uplift data was collected using an iPhone 15 with the slow motion video recording feature at 240 frames per second. The device was placed in a tripod with the athlete, force plate setup, and sufficient overhead space in view to capture all jumps in frame.

We applied Uplift’s proprietary keypoint detection model to identify major anatomical landmarks in the videos. The model returns 3D (X/Y/Z) positions of each keypoint over the duration of the recording. The keypoints collectively represent how a person’s body moves during the activity of interest. We use keypoint locations to identify takeoff and landing events, constituting jump flight time and thus, jump height. The equation to estimate jump height from flight time is: h = t 2 g/8 , with h as jump height (m), t as flight time (s) and g is the gravity acceleration (9.81 m/s 2 ).[1,2,3] This is a common method to estimate jump height, particularly for video and force-based systems.

Data Analysis

We performed regression analysis to examine the relationship between the outputs for jump height, reactive strength index and flight time. We calculated mean absolute error to examine the errors between VALD and Uplift estimates.

Results

We collected a total of 51 valid trials across 10 elite-level baseball athletes.

Jump Height

Uplift estimated jump height with a mean absolute error of 0.61 inches and 95% variance explained (coefficient of determination) compared to VALD Force Decks.

Flight Time

Because both Uplift and VALD determine jump height from flight time, the results for flight time are nearly identical to those from jump height, with 95% variance explained and a mean absolute error of 0.0096 seconds between the two systems.

RSI

Uplift has moderate agreement with VALD on RSI-Mod, with a mean absolute error of 0.079 (unitless) and with 58% variance explained between the two measures.

Discussion

Overall, Uplift single camera biomechanical analysis showed very high agreement for jump height and flight time compared to VALD Force Decks. Uplift’s estimate of RSI-Mod demonstrated moderate agreement with VALD Force Decks.

Both measurement methods have their benefits and drawbacks, as they work in different ways. Uplift uses video based recording, identifying anatomical keypoints, fitting those to a biomechanical model, and using the 3D kinematic analysis to estimate jump metrics of interest. VALD relies on direct measurement of forces under the feet during a movement.

Uplift’s method generalizes to broad kinematic analysis, as standard metrics will soon be available for countermovement jumps (such as: lower body joint angles, landing risk metrics for knee joint health, and segmental angular velocities). However, Uplift records at a maximum of 240 Hz, thus analyzing very fast movements with fine timing precision can be more challenging. This may be a partial explanation for why Uplift RSI-Mod measures do not agree as closely with VALD, which samples at 1000 Hz.

Another reason for the difference in RSI-Mod may be due to inconsistencies on how athletes pre-load their countermovement jump. Arm swing and small vertical movements of the body will directly influence ground-level forces, but may not impact pelvis keypoint vertical position. This difference would impact jump contraction time, and therefore influence RSI-mod. If ground-level forces change before the pelvis keypoint moves, then we may assume that VALD contraction time would be longer than Uplift’s, and thus, result in a slightly lower RSI-mod estimate, as we see in the results.

Both VALD and Uplift show strong agreement on measures of jump height and flight time. While VALD’s system is well-suited for environments with stable, flat ground and limited space, Uplift is better equipped for larger, variable spaces. As additional movements assessments are added to the Uplift app, beyond countermovement jumps, it will be important to continue to explore how the kinematic measures, like joint angles and velocities, can add to the comprehensive movement insights to inform athletes and coaches.