Active Urban Sensing

Workshop | Campus Sensing - Powered by Multi-dimensional Imagery Data, GIS, and Deep Learning

In this workshop, participants explored and realized the scientific exploration and analysis of Tongji University campus by using multi-dimensional imagery data, GIS and deep learning technology.

 

In recent years, with the rapid development of deep learning technology and computer vision, the analysis of emerging cities based on imagery data has emerged continuously. However, most of the traditional research focuses on open source data such as commercial street scenes and general resolution remote sensing images because of the limited use of imagery data. Students collected more detailed, real-time and complete data for Tongji campus by using mobile sensing and fixed sensing sensor equipment, and made a systematic analysis by combining different algorithm models of computer vision, including the evaluation of low-carbon renovation of campus building roofs and facades based on drones, the correlation between campus walking environment and its relaxation effect, the campus space vitality modeling of outdoor environment and crowd tracking, and the evaluation of campus mobile phone screen exposure perception through wearable cameras.

 

In the first half of the workshop, the instructors introduced the application of various sensor devices to the participants, including GoPro (mobile sensing), wearable camera (mobile/fixed mixed sensing), hunting camera (fixed sensing) and unmanned aerial vehicle (aerial sensing), and shared and introduced the relevant research results of BCL, so that the students had a deeper understanding of how to use imagery data for research.

 

In the second half of the workshop, under the guidance of the instructor, the students used QGIS and other software to make spatial statistics on the acquired data, and identified and scanned the previously acquired image data through algorithms such as Pspnet/Maskrcnn/Yolo (Image Segmentation, Detection and Classification), and made several rounds of reports and discussions on the research topics and progress, thus gaining a basic understanding and practice on how to use various perceptual modes to obtain new data and research methods.

 

After participating in this workshop, students have an overall understanding and practical experience on multi-dimensional perception, the relationship between computer vision AI and urban research, and various sensing devices.

Download
Campus Sensing final.pdf
Adobe Acrobat Document 16.3 MB

Predicting highly dynamic traffic noise using rotating mobile monitoring and machine learning method

Traffic noise, characterized by its highly fluctuating nature, is the second biggest environmental problem in the world. Highly dynamic noise maps are indispensable for managing traffic noise pollution, but two key difficulties exist in generating these maps: the lack of large amounts of fine-scale noise monitoring data and the ability to predict noise levels in the absence of noise monitoring data. This study proposed a new noise monitoring method, the Rotating Mobile Monitoring method, that combines the advantages of stationary and mobile monitoring methods and expands the spatial extent and temporal resolution of noise data. A monitoring campaign was conducted in the Haidian District of Beijing, covering 54.79 km of roads and a total area of 22.15 km2 , and gathered 18,213 A-weighted equivalent noise (LAeq) measurements at 1-s intervals from 152 stationary sampling sites. Additionally, street view images, meteorological data and built environment data were collected from all roads and stationary sites. Using computer vision and GIS analysis tools, 49 predictor variables were measured in four categories, including microscopic traffic composition, street form, land use and meteorology. Six machine learning models and linear regression models were trained to predict LAeq, with random forest performing the best (R2 = 0.72, RMSE = 3.28 dB), followed by K-nearest neighbors regression (R2 = 0.66, RMSE = 3.43 dB). The optimal random forest model identified distance to the major road, tree view index, and the maximum field of view index of cars in the last 3 s as the top three contributors. Finally, the model was applied to generate a 9-day traffic noise map of the study area at both the point and street levels. The study is easily replicable and can be extended to a larger spatial scale to obtain highly dynamic noise maps. 
Download
Zhang et al 2023 ER_RotatingSensing.pdf
Adobe Acrobat Document 9.4 MB