Real-Time Population Density Estimation in Dubai Through Deep Learning and Car Counting in Satellite Images

Accurately estimating population density is a crucial component of policy-making for the development of any country. Traditionally, population density has been estimated through labor-intensive surveys that can be time-consuming and prone to error. Census data, while useful, is only collected once every 10 years or so and can take a long time to process, depending on the geography and population of the region. This makes it difficult for organizations that require up-to-date population density information for instant policy designing. To address this issue, we propose a novel approach to estimate population density using satellite imagery. Our method leverages the correlation between car density and population density. Specifically, we validate this assumption by counting cars over Dubai city using a Faster RCNN object detector with a ResNeXt-101 (32× 8d)-FP backbone and calculating the correlation between car density and population density. Our results show a significant value of the Pearson correlation coefficient, demonstrating a strong relationship between population density and car density. This innovative approach allows for the rapid estimation of population density, without the need for time-consuming and labor-intensive surveys.