Awesome

Deep Learning Based Large-Scale Automatic Satellite Crosswalk Classification

Rodrigo F. Berriel, André Teixeira Lopes, Alberto F. de Souza, and Thiago Oliveira-Santos

IEEE Geoscience and Remote Sensing Letters: 10.1109/LGRS.2017.2719863

High-resolution satellite imagery have been increasingly used on remote sensing classification problems. One of the main factors is the availability of this kind of data. Even though, very little effort has been placed on the zebra crossing classification problem. In this letter, crowdsourcing systems are exploited in order to enable the automatic acquisition and annotation of a large-scale satellite imagery database for crosswalks related tasks. Then, this dataset is used to train deep-learning-based models in order to accurately classify satellite images that contains or not zebra crossings. A novel dataset with more than 240,000 images from 3 continents, 9 countries and more than 20 cities were used in the experiments. Experimental results showed that freely available crowdsourcing data can be used to accurately (96.78%) train robust models to perform crosswalk classification on a global scale.

Dataset Automatic Acquisition and Annotation

To download the dataset, you should run the command below for each region of interest. Be careful with your API quota.

python crosswalk-downloader.py --region={REGION_NAME} --negative --positive --key={API_KEY}
# e.g. to download the crosswalks only of the regions in Asia
python crosswalk-downloader.py --region=asia --positive --key={API_KEY}

Test with your data

Pre-trained models are available here.

This Python notebook may help you with the inference process.

Dataset

The dataset used in this work is defined by a group of city-based regions. As stated in the paper, "even though each part of the dataset is named after a city, some selected regions were large enough to partially include neighboring towns". The regions can be seen in the file regions.json and a summary of the dataset can be seen below.

Positive Samples

Negative Samples

BibTeX

@article{berriel2017grsl,
    Author  = {Rodrigo F. Berriel and Andre T. Lopes and Alberto F. de Souza and Thiago Oliveira-Santos},
    Title   = {{Deep Learning Based Large-Scale Automatic Satellite Crosswalk Classification}},
    Journal = {IEEE Geoscience and Remote Sensing Letters},
    Year    = {2017},
    DOI     = {10.1109/LGRS.2017.2719863},
    ISSN    = {1545-598X},
}