Christian Keilstrup Ingwersen

Hi, I'm am an Industrial PhD at DTU supervised by Prof. Anders Bjorholm Dahl , Assistant Prof. Morten Rieger Hannemose and Industrial Postdoc Janus Nørtoft Jensen. My research focuses on the field of 3D human pose estimation for sports applications in collaboration with the Danish company, TrackMan A/S.

With a solid background in computer vision and machine learning, I've developed a deep understanding of the latest techniques in the field. My passion for research and innovation has led me to work on some of the most challenging problems in the domain, such as real-time 3D pose estimation of athletes in various sports. I have also worked on the development of a novel 3D pose dataset for sports applications, which is the first of its kind. For a full list of publications, please see below.

Email  /  CV  /  Github  /  LinkedIn  /  Google Scholar  / 

Automated skill assessment in sports using video-based analysis holds great potential for revolutionizing coaching methodologies. This paper focuses on the problem of skill determination in golfers by leveraging deep learning models applied to a large database of video recordings of golf swings. This work contributes to the development of AI-driven coaching systems and advances the understanding of video-based skill determination in the context of golf.

With SportsPose we introduce a large marker-less 3D pose dataset, specifically targeted at sports movement. To see more visit our project page. The paper is accepted for an oral presentation at the 2023 CVPR workshop, Computer Vision in sports

We investigate current state-of-the art monocular 3D human pose estimation methods ability to predict temporally consistent poses in a domain with high frequency movements. Our investigation is based on accurate marker based motion capture data, with synchronized video of athletes performing golf swings. When qualitatively inspecting the methods estimated 3D joint locations, and projecting them into the image, the results look convincing. However, by quantitatively comparing the results to the motion capture data, we see that the model errors are significant, and too erroneous to be used for any kinematic analysis of the movements. The paper is accepted for an oral presentation at NLDL 2023.

Extension of the sparse model of MeshCNN I developed as a research intern at 3Shape. In this paper we illustrate how the model allows us to segment the left atrial appendage from the heart in a 3D reconstruction of a heart. The work was presented as a poster presentation at Northern Lights Deep Learning Conference 2022.

We presented a fully automated solution for focus calibration of a photopolymerization system. The paper was presented as an oral presentation at ASPE/euspen Summer Topical Meeting on Advancing Precision in Additive Manufacturing.