BEVT-tracker

Test passed

Matlab implementation of Boundary Effect-Aware Visual Tracking for UAV with Online Enhanced Background Learning and Multi-Frame Consensus Verification (BEVT tracker).

Publication and Citation

This paper has been published by IROS2020.

You can find this paper here: https://ieeexplore.ieee.org/document/8967674.

Please cite this paper as:

@INPROCEEDINGS{8967674,

author={C. {Fu} and Z. {Huang} and Y. {Li} and R. {Duan} and P. {Lu}},

booktitle={Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems},

title={Boundary Effect-Aware Visual Tracking for UAV with Online Enhanced Background Learning and Multi-Frame Consensus Verification},

year={2019},

volume={},

number={},

pages={4415-4422},}

Instructions

1. Download VGG-Net-19 by cliking here and put it in /model.
2. Download matconvnet toolbox here and put it in /external.
3. Configure the data sequence in configSeqs_demo_for_BEVT.m.
4. Run BEVT_Demo_single_seq.m

Note: the original version is using CPU to run the whole program. If GPU version is required, just change false in the following lines in run_BEVT.m to true:

global enableGPU;
enableGPU = false;

vl_setupnn();
vl_compilenn('enableGpu', false);

Acknowledgements

The parameter settings are partly borrowed from BACF and SRDCF paper and convolutional feature extraction function is borrowed from HCF.

Results

The following are the results from the experiment conducted on 100 challenging sequences extracted from UAV123@10fps.