KineMo Accuarcy Peer Reviewed Publication

We are incredibly excited to announce the acceptance and publication of our first peer reviewed paper, which you can find here, on exercise quantification from a single camera. We cannot understate the importance to the KineMo team of backing up what we say we can do with academic rigour and external validation. We have a pipeline of further papers planned, which include the results of studies completed in 2023 with key sports on the use and scalability of the KineMo platform.The current paper sets out requirements and processes to compute exercise metrics directly from video which are sufficiently accurate to be actionable by coaches and physiotherapists.The KineMo platform builds on this technology to accurately and objectively track and record movement competency over time during athlete rehabilitation and/or development. With KineMo you can now track kinematics across ALL your athletes ON DEMAND and over time, removing subjectivity and bias.Exercise quantification from single camera view markerless 3D pose estimation. Read the full paper below:

The industry gold standard is marker-based motion capture (e.g. Vicon, Qualisys...). These are highly accuarte systems, validated against X-rays and advanced medical imaging techniques. They require anywhere from 6 to potentially 200 high-end infrared cameras, which will track reflective markers on a person, animal or object. These systems although accurate are expensive, time-consuming to use and require a specialised lab environment with technical staff.

KineMo uses its own in-house Vicon System, both to train its Deep Learning Pose Estimation models and validate their accuracy. On motion capture days, we will set up an array of different types of cameras, from standard 30 fps low-end phone cameras to GoPros, to the newer generation phones that can achieve 120fps. These phones will record alongside the Vicon system, to track the athletes doing an array of exercises.

After the capture days, the videos will be run through KineMo and will be synced to the Vicon outputs. Instead of using the standard MPJPE for error measures, which are not relevant to this application, as it measures the offset between the joint positions, KineMo calculates the error in terms of the metrics that it uses in the app, such as Torso Angle or Knee Valgus. Once these are measured in both Vicon and KineMo, the RMSE (Root-Mean-Squared Error) is measured. This is done across all cameras, for all subjects, divided into exercises and across all metrics.

KineMo has been developed with the help of physios to aid physios. Currently, the accuracy of physios by eye assessment has been investigated in a study where 34 physios tracked squat exercises and hand-over-head forward bent functional movements. This study concluded that physios were only able to track changes with an accuracy of 12 degrees or more. These trained eye assessments are a subjective measurement, and vary widely between different professionals, rendering it almost impossible to track over time. With that in mind, KineMo demonstrates an accuracy to be easily within 12 degrees from the RMSE calculations, for all the metrics across all the exercises, and aims to continue pushing its limits to be well within 6 degrees across the whole matrix. The following table demonstrates KineMo's current validation matrix, and as we continue to develop more models, and gather more data, these error values will continue to reduce.