flyers / drone-tracking

DTB70 -- A Drone Tracking Benchmark

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Visual object tracking for unmanned aerial vehicles: A benchmark and new motion models

Created by Siyi Li and Dit-Yan Yeung at HKUST.


Drone Tracking Benchmark (DTB70) is a unified tracking benchmark on the drone platform. In this benchmark, we provide an extensive study of the state-of-the-art trackers and their various motion model variants on the DTB70 dataset. Detailed description of the benchmark can be found in our paper.


If you are using this code in a publication, please cite our paper.

    title={Visual object tracking for unmanned aerial vehicles: A benchmark and new motion models},
    author={Li, Siyi and Yeung, Dit-Yan},
    booktitle = {AAAI},


For the installation of specific trackers, please refer to the corresponding documentation.

Download dataset

Download the dataset from Baiduyun link. Put the unzipped file under the data directory. Also, change the dataset path config in file experiments/util/configDTBSeqs.m.

The dataset format follows OTB50.

How to install individual trackers

  • DAT tracker is ready to run.
  • DSST, HOG_LR, KCF, MEEM, SRDCF all need to compile mex files. Just use the compilation script in the corresponding directories.
  • SODLT and MDNet are deep learning based trackers. Please refer to the detailed documentation.

Run demo examples

Run the run_demo.m script.

Run evaluation toolkit

Run the main_running.m script under the experiments directory. You can config the trackers list in file experiments/util/configTrackers.m.

For any problems, feel free to propose issues or contact the author


DTB70 -- A Drone Tracking Benchmark


Language:MATLAB 36.9%Language:C++ 30.7%Language:C 22.0%Language:Cuda 9.2%Language:Shell 0.6%Language:Makefile 0.5%Language:M 0.1%Language:Mathematica 0.0%