2016年欧洲计算机视觉会议部分论文

and ALOV300++ [39]. Since the performance of a tracker is measured against

these benchmarks, it is critical that a holistic set of real-world scenarios and a

distribution of tracking nuisances (e.g. fast motion, illumination changes, scale

changes, occlusion, etc.) are properly represented in the annotated dataset. The

benchmark also plays a critical role in identifying future research directions in

the field and how to design more robust algorithms. What is currently lacking

in these well established benchmarks is a comprehensive set of annotated aerial

datasets that pose many challenges introduced by unmanned airborne flight.

vision capabilities (e.g. tracking, object/activity recognition, etc.) is becoming

B. Leib e et al. (Eds.): ECCV 2016, Part I, LNCS 9905, pp. 445–461, 2016.

scene. Since current benchmarks are pre-recorded scenes, they cannot provide

realistic simulator to render real-world environments and a variety of life-like

moving targets typically found in unmanned aerial recordings. The simulator

uses the Unreal Engine 4 to directly feed image frames to trackers and retrieve

tracking results to update UAV flight. Any tracker (e.g. written in Matlab or

C++) can be tested on the simulator across a diverse set of photo-realistic simu-

lated scenarios. Using this simulator enables the use of new quantitative methods

for evaluating tracker performance in the aforementioned aerial feedback loop.

fully annotated high-resolution dataset of 123 aerial video sequences comprising

more than 110 K frames. It is as large or larger than most recent, generic object

performance of state-of-the-art trackers running within its environment. The

combination of the simulator with an extensive aerial benchmark provides a

more comprehensive evaluation toolbox for modern state-of-the-art trackers and

opens new avenues for experimentation and analysis.

availability of annotated datasets specific to UAVs in which trackers can be rigor-

ously evaluated for precision and robustness in airborne scenarios. Existing anno-

tated video datasets include very few aerial sequences [42]. Surveillance datasets

such as PETS or CAVIAR focus on static surveillance and are outdated. VIVID

they do not introduce videos with many tracking nuisances addressed in this

paper and common to aerial scenarios.

trackers such as OAB [11]andIVT[38] as baselines and the best-performing

recent trackers according to [42]: Struck [13], CSK [17], ASLA [19], and TLD

[21]. In the selection process, we reject very slow trackers despite their perfor-

mance [3,4,44–47]. In addition, we include several of the latest trackers such as

Fig. 2. Column 1 and 2: Proportional change of the target’s aspect ratio and bounding