Feature Extraction Strategy for Runtime of MOT Çoklu Nesne Takip Hızı için Öznitelik Çıkarma Stratejisi

2024-01-01
Bayar, Emirhan
Aker, Cemal
Matching appearance features of objects is one of the key components of the recent Multiple Object Tracking methods. However, the computational cost of feature extraction results in extremely low processable Frames per Second (FPS) when executed on resource-constrained hardware as Edge Artificial Intelligence. This study proposes a method that selects the detections that need feature extraction on the fly and extracts features from only those detections. Applying the method to the StrongSORT algorithm, the possibility of maintaining the advantages of feature extraction while increasing FPS is discussed. Source Code: github.com/emirhanbayar/SIU_StrongSORT.
32nd IEEE Conference on Signal Processing and Communications Applications, SIU 2024
Citation Formats
E. Bayar and C. Aker, “Feature Extraction Strategy for Runtime of MOT Çoklu Nesne Takip Hızı için Öznitelik Çıkarma Stratejisi,” presented at the 32nd IEEE Conference on Signal Processing and Communications Applications, SIU 2024, Mersin, Türkiye, 2024, Accessed: 00, 2024. [Online]. Available: https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85200881759&origin=inward.