Multiple object tracking is an important task for airport gound surveillance. However, experiments show that algorithms that perform well on traditional pedestrian datasets (e.g., MOT17) do not perform well in airport scenes. This is because most multiple object tracking algorithms tend to focus more on tracking pedestrians, but airplanes have specific characteristics, such as similar size, color, and appearance, which pose a challenge to algorithms in re-id task. Additionally, the perspective in airport scenes is also significantly different from traditional pedestrian tracking datasets. To address these issues, we have created an airport scene surveillance video tracking dataset, called AGVS-T22 (Airport Ground Video Surveillance benchmark - Tracking22). AGVS-T22 contains 37 airport videos, nearly 120,000 frames, and ground truth. Each video contains some challenges in airport scene surveillance, such as occlusion, illumination changes, and multi-scale objects. We have benchmarked several state-of-the-art trackers on the dataset and observed a significant drop in performance compared to existing benchmarks. The dataset can be downloaded from www.agvs-caac.com.
AGVS-T22: A New Multiple Object Tracking Dataset for Airport Ground Video Surveillance
24.09.2023
7743971 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Object tracking in video-surveillance
British Library Online Contents | 2009
|Online Contents | 1994
British Library Online Contents | 1994
Airport surface surveillance based on video images
IEEE | 2005
|