3D Objects' Discovery Through Tracking Ortamdaki 3-Boyutlu Nesnelerin ?Izlenerek Ke sfi
33rd IEEE Conference on Signal Processing and Communications Applications, SIU 2025, İstanbul, Türkiye, 25 - 28 Haziran 2025, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu66497.2025.11112337
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: Environment semantics, object detection and recognition, multi-object tracking
- Boğaziçi Üniversitesi Adresli: Evet
Özet
In this paper, the focus is on the problem of discovering 3D objects for the semantic analysis of an environment. These objects are essentially object candidates, which are class-independent entities that can be extracted without requiring any prior knowledge about their appearance or category. It is assumed that the environment is perceived using an RGB-D camera capable of moving and looking around. A novel approach is proposed for detecting 3D object candidates present in the environment from sequential RGB-D data. In this approach, after identifying 3D object candidates in each RGB-D image, these candidates are tracked across consecutive frames to determine the objects present in the environment. Various visual and geometric relational features are used for tracking. As demonstrated in the experimental evaluations, utilizing temporal 3D data enables a more reliable and comprehensive discovery of objects.