ID 2417: Pose Estimation in Diving

Symbolic picture for the article. The link opens the image in a large view.

Performance in competitive diving is dependent on the position of the body during different key moments of the exercise. Therefore, it is important for athletes to receive immediate feedback regarding their pose at these key moments, such that they can improve if necessary. Currently, a manual approach is used to assess these parameters of interest, which is time-consuming and delays the feedback to the athletes. In movement analysis, different algorithms have been developed for automatic pose estimation. Originally, these algorithms required the athlete to wear markers at locations of interest, which affects their performance. In recent years, the use of deep learning has significantly improved the performance of markerless pose estimation. This method can then be used to extract kinetic and kinematic parameters of interest (semi-)automatically from video images. Markerless motion capture is especially useful in sports, since it is unobtrusive and does not disturb athletes. Furthermore, recent developments even allow the reconstruction of a three dimensional movement from a single camera, even when cameras are moved during the capture. However, its application in sports is challenging due to the movement speed, self-occlusion, and the fact that some poses are very athletic and therefore outside the normal motion range, which is what deep learning models are based on.


The goal of this research project is to develop a pose estimation method to identify and track diving performance, as well as investigate different relationships relating body pose to performance.

  • Estimate kinematic and kinetic variables from single camera videos of diving
  • Investigate relationship between performance and kinematic and kinetic variables
  • Focus on usability: real-time and user friendly
  • Collaboration with Olympiastützpunkt Berlin
  • Can be offered as internship / project / Bachelor’s thesis (5 or 10 ECTS project)


  • Literature study into different 2D and 3D pose estimation approaches
  • Implementation and comparison of suitable 2D and 3D pose estimation algorithms
  • Development of automatic identification of water entry
  • Data processing of experiment to extract performance parameters of interest


  • Knowledge on pose estimation


Alexander Weiß, M. Sc.

Researcher & PhD Candidate

Prof. Dr. Anne Koelewijn

Junior Professorship for Computational Movement Science

Please use the application form to apply for the topic. We will then get in contact with you.