Unbiased Mobility Data Project
What is it about?
Academic communities and public sectors have been using mobility and traffic data to address social issues (i.e GDP prediction, environmental impacts, crime, etc). One method to measure mobility and traffic state is by analyzing a series of pictures and footage from traffic cameras installed at fixed locations. However, more often than not, biases arise throughout different stages of the pipeline, from the computer vision model used to label the raw image data to the sampling bias of cameras' installed locations. This project seeks to research and investigate the biases and refine the model to improve data accuracy while utilizing the data to perform co-variate analysis in covid positivity rate, business recovery, and other socio-economic factors.
Tech Stack
- DataScienceForSocialGood
- ComputerVision
- OpenCV
- Shiny
- RStudio
My Role
Research
Working as a research fellow student, I did massive research in various topics we wanted to address in the project, ranging from preferential sampling and linear interpolation of geo-spatial data, image pre-processing methodologies to correct lighting and fidelity issues for object detection, as well as co-variate analysis with two-sample t-tests, chi-squared tests, and so forth.
Development
To aggregate and address all the research findings, we invented a Shiny app solution to visualize the traffic data and business locations in different dimensions(i.e. time, vehicle type) in the City of Surrey, perform comparisons of vehicle count data between same-intersection and nearest-neighbour cameras, and use the corrected vehicle count data from the refined model.
Please check out this article for more details about the whole experience!
Live Demo
Official Documentation
Read the full documentation📋