Skip to content

The tutorials, datasets and source codes of the crosswalk detection (zebra crossing detection) network, which is robust in real scenes and real-time in Jetson nano. cross. detect. pedestrian.

Notifications You must be signed in to change notification settings

zhangzhengde0225/CDNet

Repository files navigation

Stars Open issue Datasets Datasets

English | 简体中文

Please star this project in the upper right corner and cite this paper blow if this project helps you.

CDNet

This repository is the codes, datasets and tutorials for the paper "CDNet: a real-time and robust crosswalk detection network on Jetson nano based on YOLOv5".

CDNet (Crosswalk Detection Network) is a specific implementation of crosswalk (zebra crossing) detection and vehicle crossing behavior analysis under the vision of vehicle-mounted camera.

GA

Fig.1 Graphical abstract.

Highlights

  • A crosswalk detection and vehicle crossing behavior detection network is realized.
  • The accuracy and speed exceed YOLOv5 in the specific task.
  • High robustness in real complex scenarios such as in cloudy, sunny, rainy and at night is achieved.
  • Real-time detection (33.1 FPS) is implemented on Jetson nano edge-computing device. +The datasets, tutorials and source codes are available on GitHub.

Contribution

  • SENet (Squeeze-and-Excitation Network), F1 score up, speed slightly down
  • NST (Negative Samples Training), F1 score up, speed invariant
  • ROI (Region Of Interest), F1 score down, speed up
  • SSVM (Slide receptive field Short-term Vectors Memory), transfer crosswalk detection task into vehicle crossing behavior task, F1 score up, speed invariant
  • SFA (Synthetic Fog Augment), dataset augment, adapt to foggy weather, F1 score up, speed invariant

Installation

Get CDNet code and configure the environment, please check out docs/INSTALL.md

Model Zoo

Please check out docs/MODEL_ZOO.md

Datasets

Download trainsets and testsets, please check out docs/DATASETS.md

Quick Start

Train

Once you get the CDNet code, configure the environment and download the dataset, juse type:

python train.py --trainset_path </path/to/trainset/folder>
(such as: /home/xxx/datasets/train_data_yolov5_format) 

The training results and weights will be saved in runs/expxx/ directory.

The main optional arguments:

--device "0, 1"  # cpu or gpu id, "0, 1" means use two gpu to train.
--img-size 640 
--batch-size 32 
--epochs 100 
--not-use-SE  # use original YOLOv5 which not SE-module embedded if there is the flag

Inference

Detect the crosswalk image by image and analyze the vehicle crossing behavior.

python detect.py

The main optional arguments:

--source example/images  # images dir
--output example/output  # output dir
--img-size 640  # inference model size
--device "0"   # use cpu or gpu(gpu id)
--plot-classes ["crosswalk"]  # plot classes
--field-size 5  # the Slide receptive field size of SSVM 
--not-use-ROI  # not use roi for accelerate inference speed if there is the flag
--not-use-SSVM  # not use ssvm method for analyse vehicle crossing behavior if there is the flag

For more details, please refer to docs/INSTALL.md and docs/DATASETS.md.

Fogging Augment

If you want to augment datasets by synthetic fog algorithm, just run:

python fog_augment.py

For more details, please view the source code in fog_augment.py and /scripts/synthetic_fog.py

Results

Results

Fig.2 Performance compared to YOLOv5.

CDNet improves the score for 5.13 points and speed for 10.7 FPS on Jetson nano for detection size of 640 compared to YOLOv5.

For detection size of 288, the improvements are 13.38 points and 13.1 FPS.

Contributors

CDNet is authored by Zhengde Zhang, Menglu Tan, Zhicai Lan, Haichun Liu, Ling Pei and Wenxian Yu.

Currently, it is maintained by Zhengde Zhang (drivener@163.com).

Please feel free to contact us if you have any question.

The Academic homepage of Zhengde Zhang: zhangzhengde0225.github.io.

Acknowledgement

This work was supported by the National Natural Science Foundation of China [Grant Numbers: 61873163].

We acknowledge the Center for High Performance Computing at Shanghai Jiao Tong University for providing computing resources.

We are very grateful to the yolov5 project for the benchmark detection algorithm.

We are very grateful to the tensorrtx project for the deployment techniques to the Jetson nano.

Links

Detect Video Samples:https://www.bilibili.com/video/BV1qf4y1B7BA

Read Full Text of This Paper:https://rdcu.be/cHuc8

Download Full Text of this Paper:https://doi.org/10.1007/s00521-022-07007-9

Project Introduction on CSDN:http://t.csdn.cn/Cf7c7

If it is helps you, please star this project in the upper right corner and cite this paper blow.

Citation

@article{CDNet,
author={Zheng-De Zhang, Meng-Lu Tan, Zhi-Cai Lan, Hai-Chun Liu, Ling Pei and Wen-Xian Yu},
title={CDNet: a real-time and robust crosswalk detection network on Jetson nano based on YOLOv5},
Journal={Neural Computing and Applications}, 
Year={2022},
DOI={10.1007/s00521-022-07007-9},
}

License

CDNet is freely available for free non-commercial use, and may be redistributed under these conditions. For commercial queries, please drop an e-mail at drivener@163.com. We will send the detail agreement to you.

About

The tutorials, datasets and source codes of the crosswalk detection (zebra crossing detection) network, which is robust in real scenes and real-time in Jetson nano. cross. detect. pedestrian.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages