A Multiple Aircraft Tracking Dataset for Airport Traffic Surveillance

Multiple Aircraft Tracking (MAT) is the foundation of many traffic safety applications in airports. However, experiments show that the state-of-the-art algorithms in Multiple Object Tracking (MOT) deteriorate significantly in the airport scene, and the performance drop can even reach 40%. This is because aircraft and airport scenes possess unique characteristics. For instance, aircraft’s low-compact design causes significant changes in appearance from different angles, while the expansive nature of airports presents challenges in multi-scale issues, particularly at smaller scales. In this paper, we introduce a new dataset Airport Ground Video Surveillance-Tracking24 (AGVS-T24), which could servers as a benchmark to study the challenges in MAT. AGVS-T24 includes 53 airport videos, totaling more than 150,000 frames, and precise manual annotations. AGVS-T24 comprehensively presents various motion patterns of the aircraft, such as takeoff, landing, docking, undocking, taxiing, turning, acceleration, deceleration, etc. AGVS-T24 also contains a variety of MAT challenges, such as appearance change, simultaneous multi-scales, weather and illumination changes, similar appearance between aircraft, as well as tracking in infrared and panoramic modes, and so on. Furthermore, we conduct a simple review of current MOT algorithms, and 20 classic algorithms are selected and tested on AGVS-T24. Finally, we also summarize some principles for designing MAT algorithms. This dataset can be downloaded fromwww.agvs-caac.com/AGVS-T/agvst24.html