Train Like A Machine
In this project, students explore how sensors and machine learning algorithms can be used to analyze sports exercises. Students work as teams on tasks they set themselves and in coordination with the project supervisors. Topics include, among others, the process of data preprocessing, the segmentation or labeling of data as well as feature engineering and classification.
To date, the project has taken place in WiSe 2021, SoSe 22, WiSe 2022, SoSe 2023, and SoSe 2024 (in English for the first time). André and Jan are happy to answer your questions!
Some projects realised during the semesters:
Group „DailyChallenge“
With the help of IMU sensors, squats were assessed in real time. The assessment was realized using the Dynamic Time Warping algorithm.
Group „Bauch-Beine-Po“
Using IMU sensors, an algorithm was developed to determine how many repetitions a person can still perform during a barbell exercise (repetitions in reserve).
Group „SmartUp“
Based on IMU sensors, the focus was placed on the detection of correctly executed pull-ups.
Group „Boxingmachine“
The goal was to classify correctly and incorrectly executed straight punches (Krav Maga) as such. IMU sensors were also used by this group.
Group „Tennisarm“
A table tennis serve detection system was implemented using an IMU sensor and machine learning.
Bachelor Theses
Graph Convolutional Network based Gesture Interpretation of Skeleton Data
Thore Ottenheym, 2024, Link
Effects of Image Resolution on Skin Lesions Classification with Residual Neural Networks
Niklas Hoefflin, 2024, Link
Human joint angle measurement exploiting kinematic constraints using an inertial motion capture system
Ihor Zhvanko and Nataliya Didukh, 2024, Link
Algorithmic activity detection based on motion and position sensors
Phillip Schackier, 2021, Link
People
André Jeworutzki
