Objective: This project aims towards the development and validation of a fully automated framework for tremor analysis using common video recordings.

Background: Till date, sensor-based tremor analysis is an essential tool for the quantification of tremor. Recent studies have shown promising results for Computer Vision (CV) based approaches of tremor assessment. Our approach could provide a fully automated analysis of upper limb tremor and incorporate a three-dimensional assessment via state-of-the-art depth estimation on 2D-videos.

Method: Videos were recorded using the build in webcam of a 13’’ Macbook^tm pro 2021 at 30Hz with full HD resolution. Each video was 30 seconds long with an IMU (Adafruit LSM6DSO32, 3 gyroscopes, ±125 dps, 208 Hz) recording synchronously. For computer vision hand recognition we relied on the “Mediapipe” framework by Google ^tm research. For signal processing, the the python “Scipy”-package was used.

Results: In total 4 subjects with different types of tremor were included. The difference of the detected tremor peak frequencies identified with both approaches covered a range of 0.36 Hz-0.51 Hz.

Conclusion: Our CV-based tremor analysis shows a good preliminary accuracy compared to the IMU based analysis. This technique holds the potential for accurate tremor analysis right on the practitioners’ smartphone.

« Back to 2022 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/validation-of-fully-automated-tremor-analysis-on-common-videos-based-on-computer-vision/