Název:	Multiple Object Tracking by Bounding Boxes Without Using Texture Information and Optical Flow
Autoři:	Lipovits, Ágnes Czúni, László Tömördi, Katalin Vörösházi, Zsolt
Citace zdrojového dokumentu:	WSCG 2021: full papers proceedings: 29. International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision, p. 309-316.
Datum vydání:	2021
Nakladatel:	Václav Skala - UNION Agency
Typ dokumentu:	conferenceObject konferenční příspěvek
URI:	http://hdl.handle.net/11025/45037
ISBN:	978-80-86943-34-3
ISSN:	2464-4617 2464–4625(CD/DVD)
Klíčová slova:	sledování objektu;detekce objektů;Maďarská metoda;RetinaNet
Klíčová slova v dalším jazyce:	object tracking;object detection;Hungarian method;RetinaNet
Abstrakt v dalším jazyce:	Object tracking is a key task in many applications using video analytics. While there is a huge number of algo-rithms to track objects, there is still a need for new methods to solve the correspondence problem under certaincircumstances. In our article, we assume a very typical but still open scenario: a still image object detector hasalready identified the objects to be tracked; thus, we have object labels, confidence values, and bounding boxes ineach video frame captured at a low sampling rate. That is, optical flow methods difficult to be applied (also dueto bad lighting conditions, cluttered or homogeneous areas and strong ego-motion), and moreover, many objectslook similar (having the same category labels). Our proposed approach is based on the Hungarian method andincorporates the above information into the cost function evaluating the possible pairings of objects. To considerthe uncertainty of the detector, the elements of the confusion matrix also contribute to the cost of pairs, as wellas the probability of spatial translations based on prior observations. As a use case, we apply the algorithm to adata-set, where images were captured from onboard cameras and traffic signs were detected by RetinaNet. Weanalyze the performance with different parameter settings.
Práva:	© Václav Skala - UNION Agency
Vyskytuje se v kolekcích:	WSCG 2021: Full Papers Proceedings

Všechny záznamy v DSpace jsou chráněny autorskými právy, všechna práva vyhrazena.