Tracklet clustering for 2D tracking.
Please prepare the detection file with the format that follows MOT challenge https://motchallenge.net/instructions/.
Set the directory in demo.m.
Input:
img_path: the directory of the image folder that contains the video sequence.
det_path: the directory of the detection file from the previous step.
seq_name: the name of the video sequence.
ROI_path: the directory of the ROI mask. If no ROI provided, use empty matrix instead.
img_save_path: the directory of the output image after tracking.
result_save_path: the directory of the tracking result. The result follows the UA-Detrac format. https://detrac-db.rit.albany.edu/instructions.
Set the parameters in demo.m.
det_score_thresh: detection score threshold between 0 and 1.
IOU_thresh: IOU threshold for detection asscociation across frames between 0 and 1.
color_thresh: color threshold for detection asscociation across frames between 0 and 1.
lambda_time: time interval cost.
lambda_split: tracklet separation cost.
lambda_reg: smoothness cost.
lambda_color: color change cost.
lambda_grad: velocity change cost.
Use this bibtex to cite this repository:
@inproceedings{tang2018single,
title={Single-camera and inter-camera vehicle tracking and 3D speed estimation based on fusion of visual and semantic features},
author={Tang, Zheng and Wang, Gaoang and Xiao, Hao and Zheng, Aotian and Hwang, Jenq-Neng},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops},
pages={108--115},
year={2018}
}