Global-Local MAV Detection under Challenging Conditions based on Appearance and Motion

This is an official pytorch implementation of the 2024 IEEE Transactions on Intelligent Transportation Systems paper:

Global-Local MAV Detection under Challenging Conditions based on Appearance and Motion

This repository contains the basic codes for GLAD, the full codes with Kalman Filter, Adaptive Search Region, and other codes will be published in the future.

In this paper, we propose a global-local MAV detector that can fuse both motion and appearance features for MAV detection under challenging conditions.

Quick Start for GLAD

demo file for GLAD

python GLAD.py

GLAD_MC

demo file for GLAD with motion compensation

Installation

Please find installation instructions in INSTALL.md.

Train

If you want to train on your own datasets, you should re-train global YOLOv5, local YOLOv5, and the appearance-based classifier.

Dataset

• BaiduYun (code:z1xb)
• GoogleDrive

This dataset includes 60 videos and 107497 frames, the average object size is only 0.02% of the image size (1920 × 1080). The annotation files follow the Pascal VOC XML format. In addition, we provide a python code for extracting images from a video.

In this paper, 45 videos are used for model trainning and validation (randomly split with 5:1), and 15 videos are used for testing. The video ID for testing is:05, 08, 09, 10, 19, 30, 41, 43, 46, 47, 58, 63, 65, 70, 86.

Conditions
Ordinary	09	10	30	47	70
Complex	05	08	58	65	86
Small	19	41	43	46	63

If you have any problem when using this dataset, please feel free to contact: guohanqing@westlake.edu.cn.

Citing

If you find our work useful, please consider citing:

@article{guo2023globallocal,
      title={Global-Local MAV Detection under Challenging Conditions based on Appearance and Motion}, 
      author={Hanqing Guo and Ye Zheng and Yin Zhang and Zhi Gao and Shiyu Zhao},
      journal={IEEE Transactions on Intelligent Transportation Systems},
      year={2024},
      publisher={IEEE}
}