Community noise measure via crowd sourcing

• Main Author: Lim, Yun Jie
• Other Authors: Lee Bu Sung, Francis
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Social sciences::Geography::Environmental sciences; Engineering::Computer science and engineering::Data::Coding and information theory
• Online Access: https://hdl.handle.net/10356/153343
• Institution: Nanyang Technological University

Noise pollution is linked to a range of health problems that impairs health and worsens living standards, resulting in both intangible (psychological, health) and physical (financial) losses. This final year project reports on the gathering of community noise data around Singapore, the classification of the noise source, eg. Machine, traffic, human and lastly the simulation of trees and their possible noise mitigating effects. Firstly, a crowdsourcing approach was taken where participants contribute data of their noise exposures using the Noise Capture application. NoiseCapture is an application that is built upon the notions of citizen science and participatory sensing. The application employs GPS-enabled mobile phones as noise sensors to monitor the users’ noise exposure in their daily lives. The user-generated metadata and geo-localized measures are then automatically shared online with the public. Initially, noise encountered on campus was recorded and analysed. The results were promising, providing information on noise from the human perspective as they go about their daily lives in the campus environment. The study shows that the noise on campus is mainly due to human and mechanical noise. By mashing the noise data with Wi-Fi log data, the results show a good positive correlation between noise level and human density in an area. The next experiment was to analyse vehicular traffic noise. The area of study is a walking path located in a section of the Clementi area perpendicular to the West Coast Highway. Through data cleaning to reduce the noise in the data, the results show a logarithmic decrease as the distance from the highway increases. The final experiment dives deeper into how trees can help towards reducing road noise. Obtaining tree location data is important in modelling noise propagation flows. Several approaches of location data collection were analysed based on several considerations such as their cost and the reproducibility of those methods in other countries outside of Singapore. Exploretree-sg is a method that crawls tree location data from MAVEN API. MAVEN is a government platform that consolidates vegetation and biodiversity data in Singapore. However, this method is limited to Singapore and hence two other methods, Treepedia and Mapillary are explored. These two methods can be replicated in other countries. Both methods use Green View Index (GVI) to quantify the presence of trees in a geographical location point. The difference between these two methods is their source of data. Treepedia obtains image data from Google Map’s Street View Static API which is advantageous in the number of location data points but is limited by its cost. Mapillary on the other hand obtains image data from the Mapillary v4 API. The data from Mapillary is crowdsourced from vehicle cameras, making it free to use but the location of its data points is limited to roads. Since GVI only quantifies the presence of trees instead of its actual location, the experiment also looks at using TensorFlow Object Identification API to build a tree detection algorithm. Various pre-trained models obtained from TensorFlow 2 Detection Model Zoo were trained and tested and the Single-shot MultiBox Detector (SSD) with MobileNet was observed to perform the best. Using the tree location data approaches discussed, tree modelling can be carried out in the future. In addition, the method of obtaining tree location data using the tree detection algorithm created can be further explored.