VPS-Net

A vacant parking slot detection method in the around view image based on deep learning.

Requirement

We ran our experiments with PyTorch 1.0.1, CUDA 9.0, Conda with Python 3.6 and Ubuntu 16.04.

Installation

Clone and install requirements

$ git clone https://github.com/weili1457355863/VPS-Net.git
$ cd VPS-Net
$ conda create --name vps-net python=3.6
$ conda activate vps-net
$ pip install -r requirements.txt

Download pretrained weights

$ mkdir weights
$ cd weights/

Download the weights of detection network and classification network.

Download ps2.0 and PSV dataset

$ mkdir data
$ cd data/

Download the ps2.0 dataset or the PSV dataset

Test

Uses pretrained weights to detect the vacant parking slot in the around view image.

$ vps_net.py [-h] [--input_folder INPUT_FOLDER]
                  [--output_folder OUTPUT_FOLDER] [--model_def MODEL_DEF]
                  [--weights_path_yolo WEIGHTS_PATH_YOLO]
                  [--weights_path_vps WEIGHTS_PATH_VPS]
                  [--conf_thres CONF_THRES] [--nms_thres NMS_THRES]
                  [--img_size IMG_SIZE] [--save_files SAVE_FILES]

Example (ps2.0 dataset)

Test on the ps2.0 dataset. The detection results including images and files will be saved.

$ python vps_net.py --input_folder data/ps2.0/testing/all --save_files 1

Testing results

Extral annotation

In order to facilitate other researchers, the annoation for vacant parking slots of ps 2.0 and PSV datasets has been made publicly avaliable.

Citation

@article{li_vacant_2020,
	title = {Vacant Parking Slot Detection in the Around View Image Based on Deep Learning},
	volume = {20},
	doi = {10.3390/s20072138},
	pages = {2138},
	journal = {Sensors},
	author = {Li, Wei and Cao, Libo and Yan, Lingbo and Li, Chaohui and Feng, Xiexing and Zhao, Peijie},
	date = {2020-04-10}
}