├── .DS_Store ├── fig1.png ├── fig2.png ├── logo.png └── README.md /.DS_Store: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/ualsg/Visual-soundscapes/HEAD/.DS_Store -------------------------------------------------------------------------------- /fig1.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/ualsg/Visual-soundscapes/HEAD/fig1.png -------------------------------------------------------------------------------- /fig2.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/ualsg/Visual-soundscapes/HEAD/fig2.png -------------------------------------------------------------------------------- /logo.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/ualsg/Visual-soundscapes/HEAD/logo.png -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Sensing soundscapes from street view imagery 2 | 3 | ## Introduction 4 | The acoustic environment is an essential component of healthy and sustainable cities. We present a machine-learning framework for portraying large-area high-resolution urban soundscapes using ubiquitous street view imagery(SVI), without ground measurements. This dataset includes two parts: (1) SVI visual features and soundscape indicators data; (2) field-measured SVI and noise intensity data. 5 | 6 | ### SVI visual features and soundscape indicators data 7 | We also extract a total of 482 visual features from each imagery using computer vision algorithms and deep learning model. The extraction of visual features including: Object Detection(by [Faster R-CNN](https://pytorch.org/vision/stable/generated/torchvision.models.detection.fasterrcnn_resnet50_fpn.html#torchvision.models.detection.fasterrcnn_resnet50_fpn)), Semantic Segmentation(by [DeepLabV3P](https://github.com/PaddlePaddle/PaddleSeg)), and Scenes Classification(by [ResNet](https://pytorch.org/vision/stable/generated/torchvision.models.resnet50.html?highlight=resnet#torchvision.models.resnet50) 8 | ) features and low-level features by the algorithms from the [OpenCV](https://opencv.org/) library. Each imagery was labeled a total of **15 soundscape indicators** are divided into four categories: 9 | * Noise intensity 10 | * Sound quality 11 | * Sound sources: traffic noise, human sounds, natural sounds, mechanical noise, and music noise 12 | * Perceptual emotion: pleasant, chaotic, vibrant, uneventful, calm, annoying, eventful, and monotonous 13 | 14 |
15 | 16 |
17 | 18 | We invited a total of 338 people to participate in the survey via [Amazon Mechanical Turk](https://www.mturk.com/) and social media. 19 | 20 | ### Field-measured SVI and noise intensity data 21 | The devices used in the collection include a Sound Level Meter(UT353BT) for noise intensity recording and a smartphone for the shooting of videos and street view imagery. Each investigation points include: 22 | * Three-minute video clips 23 | * 4-10 street view imageries 24 | * Three-minute recording of variations in sound intensity 25 | 26 |
27 | 28 |
29 | 30 | ## Access 31 | 32 | The raw SVI data can be dowmloaded at [figshare](https://figshare.com/s/2e7e23a20be112d828ea). The SVI visual features and soundscape indicators data can be found at *Label_Features*. The Field-measured SVI and noise intensity data can be downloaded at [figshare](https://figshare.com/s/2e7e23a20be112d828ea). 33 | 34 | ## Paper 35 | 36 | A [paper](https://doi.org/10.1016/j.compenvurbsys.2022.101915) about the work was published in _Computers, Environment and Urban Systems_ and it is available open access [here](https://ual.sg/publication/2023-ceus-soundscapes/2023-ceus-soundscapes.pdf). 37 | 38 | If you use this work in a scientific context, please cite this article. 39 | 40 | Zhao T, Liang X, Tu W, Huang Z, Biljecki F (2023): Sensing urban soundscapes from street view imagery. Computers, Environment and Urban Systems, 99: 101915. doi:10.1016/j.compenvurbsys.2022.101915 41 | 42 | ``` 43 | @article{2023_ceus_soundscapes, 44 | author = {Zhao, Tianhong and Liang, Xiucheng and Tu, Wei and Huang, Zhengdong and Biljecki, Filip}, 45 | doi = {10.1016/j.compenvurbsys.2022.101915}, 46 | journal = {Computers, Environment and Urban Systems}, 47 | pages = {101915}, 48 | title = {Sensing urban soundscapes from street view imagery}, 49 | volume = {99}, 50 | year = {2023} 51 | } 52 | ``` 53 | 54 | ## License 55 | This dataset is released under the [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/) license. 56 | 57 | ## Contact 58 | Feel free to contact [Tianhong Zhao](https://ual.sg/authors/tianhong/) or [Filip Biljecki](https://ual.sg/authors/filip/) should you have any questions. 59 | For more information, please visit the website of the [Urban Analytics Lab](https://ual.sg/), National University of Singapore. 60 | 61 | ## Acknowledgements 62 | We gratefully acknowledge the participants of the survey and the input data. We thank the members of the NUS Urban Analytics Lab for the discussions. The Institutional Review Board of the National University of Singapore has reviewed and approved the ethical aspects of this research (reference code NUS-IRB-2021-906). 63 | 64 | --------------------------------------------------------------------------------