Logo des Repositoriums

Auflistung nach Schlagwort "video analysis"

1 - 2 von 2
  • Konferenzbeitrag
    The digital skeleton in modern video analysis - inter- and intraspecific comparsionofindividualrigs
    (INFORMATIK 2023 - Designing Futures: Zukünfte gestalten, 2023) Pistorius, Elena; Richter, Sabine; Labudde, Dirk
    New methods for identifying individuals based on their digital skeleton (rig) gain attention in digital forensics. To ensure a solid comparison, understanding the development of this digital skeleton is crucial. One approach is manually creating a rig using a parametrized 3D room and cameras with embedded pictures, capturing multiple angles of the same person with markers on defined spots. Using these pictures and the room, a movable marker-based rig can be built in software like Blender (https://www.blender.org). It can be used to fit the rigged doll into another scene or compare proportions. For a similar process, algorithm-based software like OpenPose (https://github.com/CMU- Perceptual-Computing-Lab/openpose) estimates a digital skeleton from pictures or videos. The output includes JSON files with joint coordinates and pictures with embedded skeletons, which can manually be processsed into a rig. While these procedure can save time and expand applications, precision and supervision differ from marker-based rigs. A comparative study assessing similarity is discussed in the following paper.
  • Konferenzbeitrag
    Potential approach for targeted matching of people in video footage based on 3D human pose estimation
    (INFORMATIK 2024, 2024) Richter, Sabine; Labudde, Dirk
    Analyzing video footage can make an important contribution to solving crimes, but it is often limited by challenges such as hidden faces. Modern methods of the digital age rely on anthropometric measurements to overcome these limitations. However, current techniques, often originating from other sectors such as the gaming industry, offer new potential. In this context, a possible new approach based on the application of 3D human pose estimation will be presented. This would offer the possibility of directly matching two people (perpetrator, suspect) using one video of each. In particular, this new approach would significantly reduce the scope of the necessary work steps and the time required. This also applies in comparison to methods that already use 2D position estimation. The new approach presented here could contribute to a significant increase in forensic efficiency if successfully implemented after testing and evaluation on a study dataset and real case data.