SAMF-TLD Based Pedestrian Multi-Target Tracking

Aiming at the tracking loss problem caused by target scale change and occlusion in pedestrian multi-target tracking, a scale adaptive kernel correlation filter with tracking-learning-detection (SAMF-TLD) based multi-target tracking method is proposed. This method combines a short-time tracker with t...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Proceedings (International Conference on Computer Engineering and Applications. Online) s. 36 - 40
Hlavní autoři: Liu, Bao, Zhao, Yuge, Yang, Haoning, Cao, Le
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 07.04.2023
Témata:
Adaptive filters
Correlation
Filtering algorithms
Lighting
multi-target tracking
multi-threaded programming
Programming
Real-time systems
scale adaptive kernel correlation filter
Target tracking
tracking-learning-detection
ISSN:2159-1288
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Aiming at the tracking loss problem caused by target scale change and occlusion in pedestrian multi-target tracking, a scale adaptive kernel correlation filter with tracking-learning-detection (SAMF-TLD) based multi-target tracking method is proposed. This method combines a short-time tracker with the traditional track-detect-learn (TLD) multi-target tracking algorithm, which improves the real-time performance of the algorithm and overcomes the tracking failure problem of the TLD algorithm under illumination change and large scale change. Secondly, by judging the confidence of the tracking results of the scale adaptive kernel correlation filter (SAMF), the problem of updating the filter template when the target is occluded is solved, therefore, in order to improve the tracking efficiency, effectively avoid the tracking failure problems caused by the target being occluded and the tracking target scale change. In the experiment, 2DMOT-15 and MOT-16 data sets were selected for tracking verification, and the common evaluation indicators (MOTA, MOTP, MP, MT, and IDS) were compared and analyzed. The results show that the proposed method is superior to the TLD pedestrian multi-target tracking algorithm in the accuracy, and the SAMF tracking algorithm in terms of anti-occlusion.
ISSN:2159-1288
DOI:10.1109/ICCEA58433.2023.10135431