It gets rather hot in our room in the afternoon. While using the aircon was an option, it was not economically feasible in the long run. We wondered if using plants as a natural cooling option was possible, and if so, where can we place it on the walls outside the building to reduce the heat from coming in.
What it does
bushier is a visual advisory tool for building owners to identify optimal vertical greenery solutions for energy and cost saving measures through ML visual overlays. Use bushier to capture / upload building profiles to our ML models, to obtain optimal overlay of vertical greenery, potential energy and cost savings.
How we built it
We built a segmentation model (HRnet + OCR head) trained on various building profile datasets, ranging from LabelMeFacade, eTRIMS to CMP Facade. Then, used IMU, location data, and other web-scrapped data to predict build temperature, incident radiation, vertical garden placement, temperature reduction, energy and cost savings. We also explored multiple ways to predict building profile area, ranging from depth estimation methods to conventional geometrical methods.
Challenges we ran into
- Getting the ML model to produce meaningful results (i.e. being able to train properly on the training sets)
- Connecting our middle end Flask with front end React Native, and other integration hell
- Finding good features / predictors of building temperature / incident radiation
Accomplishments that we're proud of
- Eventually the ML model worked and produced a sensible set of results
- We linked up the frontend with the ML model and backend through Flask
- Prediction of building temperature on a range of inputs seems sensible
- AR output seems rather attractive, compared to we expect initially
What we learned
Too many things. The last 48 hours was a whirlwind as we faced all kinds of issues and had to debug them, even such as git pull/merge issues. However, every problem we faced was a learning opportunity. Our biggest takeaway was that it takes time to learn and having failures along the way are normal. Just don't give up.
What's next for Cooling Singapore Sustainably
We hope to make the app more robust, to be able to take in more kinds of building surfaces, and produce a greater range of metrics that can be useful in encouraging vertical greenery solutions in Singapore
1) Yuan. Y, X. Cheng, X. Chen, J. Wang, Segmentation Transformer: Object-Contextual Representations for Semantic Segmentation, ECCV 2020, https://arxiv.org/abs/1909.11065
2) Ke et. al. High-Resolution Representations for Labeling Pixels and Regions, CVPR 2020, https://arxiv.org/abs/1904.04514
3) B. Frohlich, E. Rodner, J. Denzler, A Fast Approach for Pixelwise Labeling of Façade Images, ICPR 2010, https://www.researchgate.net/publication/220929012_A_Fast_Approach_for_Pixelwise_Labeling_of_Facade_Images
4) Holm, D. (1989). Thermal improvement by means of leaf cover on external walls — a simulation model. Energy and Buildings, 14(1), 19–30. https://doi.org/10.1016/0378-7788(89)90025-x
5) Canadarma, W. W., Juniwati, A., & Kristanto, L. (2006). Effect Of Orientation On Indoor Temperature Case Study: Yekape Penjaringansari Housing in Surabaya. INTA Conference 2006 - Harmony in Culture and Nature.
6) Pérez, G., Coma, J., Martorell, I., & Cabeza, L. F. (2014). Vertical Greenery Systems (VGS) for energy saving in buildings: A Review. Renewable and Sustainable Energy Reviews, 39, 139–165. https://doi.org/10.1016/j.rser.2014.07.055
7) Pan, J., Tang, J., Caniza, M., Heraud, J.-M., Koay, E., Lee, H. K., Lee, C. K., Li, Y., Ruiz, A. N., Santillan-Salas, C. F., & Marr, L. (2021). Correlating indoor and outdoor temperature and humidity in a sample of buildings in tropical climates. https://doi.org/10.31224/osf.io/6ap5q
8) Wong, N. H., Kwang Tan, A. Y., Chen, Y., Sekar, K., Tan, P. Y., Chan, D., Chiang, K., & Wong, N. C. (2010). Thermal evaluation of vertical greenery systems for building walls. Building and Environment, 45(3), 663–672. https://doi.org/10.1016/j.buildenv.2009.08.005