A Demo of Sequences in DarkTrack2021

Image credit: Unsplash

The green boxes mark the tracking objects, while the top left corner of the images display sequence names. Low brightness makes it hard to identify objects, which leads nighttime UAV tracking to an extremely challenging task.

This work constructs a large-scale nighttime UAV tracking benchmark for a comprehensive evaluation. The benchmark comprises 110 challenging sequences with 100K frames in total. All sequences are captured at nighttime in urban scenes by a DJI Mavic Air 2 UAV, with a frame-rate of 30 frames/s (FPS). This figure displays some first frames of selected sequences. The objects in DarkTrack2021 contain person, bus, car, truck, motor, dog, building, etc., covering numerous scenarios of real-world UAV nighttime tracking tasks. Therefore, a tremendous number of scenarios with various challenges, including viewpoint change, fast motion, large occlusion, low resolution, low brightness, out-of-view, etc., are involved in DarkTrack2021. In addition, scenes in the newly developed benchmark are generally captured in complex urban, where light conditions are more cluttered, frequently bringing severe illumination variation and overexposure/underexposure challenges. In that case, Dark- Track2021 provides an extensive nighttime UAV tracking performance evaluation.

Baidu Yun (code:v4rr)