Dynamic Vision: From 2D Multi-Object to 3D Point Tracking
MCML Authors
Jenny Seidenschwarz
* Former Member
Abstract
Jenny Seidenschwarz
* Former Member
Abstract
Dynamic Vision, the processing of 2D and 3D spatio-temporal data such as video or LiDAR point cloud streams, is fundamental to various applications, including smart manufacturing, autonomous systems, and mixed reality. Recently, the field has significantly improved in tasks such as tracking at various granularities, from object- to particle-scale. Many approaches integrate appearance, semantic, and motion cues to enhance performance. We introduce key concepts of Dynamic Vision before analyzing the strengths of different cues across tasks of varying granularities.
BibTeXKey: Sei25