Keyfilter-Aware Real-Time UAV Object Tracking

Matlab implementation of Keyfilter-Aware UAV Object Tracker (KAOT).

Description and Instructions

How to run

• download matconvnet
• install.m - Compile libraries and download network
• run_KAOT.m

Features

1. Deep CNN features. It uses matconvnet [1], which is included in external_libs/matconvnet/. The imagenet-vgg-m-2048network available at http://www.vlfeat.org/matconvnet/pretrained/ was used. You can try other networks, by placing them in the feature_extraction/networks/ folder.
2. HOG features.
3. Colorspace features. Currently grayscale and RGB are implemented.

Publications

@inproceedings{Li2020ICRA, title={Keyfilter-aware real-time uav object tracking}, author={Li, Yiming and Fu, Changhong and Huang, Ziyuan and Zhang, Yinqiang and Pan, Jia}, booktitle={IEEE International Conference on Robotics and Automation}, year={2020} }

@article{Li2020TMM, title={Intermittent Contextual Learning for Keyfilter-Aware UAV Object Tracking Using Deep Convolutional Feature}, author={Li, Yiming and Fu, Changhong and Huang, Ziyuan and Zhang, Yinqiang and Pan, Jia}, journal={IEEE Transactions on Multimedia}, year={2020} }

Acknowledgements

This work borrowed the feature extraction modules from the STRCF tracker (https://github.com/lifeng9472/STRCF) and the parameter settings from BACF (www.hamedkiani.com/bacf.html).