Estimation of urban traffic density from street camera images
Tutor / director / evaluatorBéjar Alonso, Javier
Document typeBachelor thesis
Rights accessOpen Access
Traffic is a very real problem in today’s world. Elon Musk, CEO of SpaceX and Tesla, has created a start-up which aims to bore tunnels to better manage traffic. While we do not dare to challenge him, we propose to better manage traffic on roads by leveraging currently available technologies in a novel way. Our project aims to give accurate real-time predictions of traffic, so that prospective commuters can choose routes that are free of traffic, thereby automatically balancing traffic. There are systems in place which give estimates of traffic but ours is cheaper, easier, and much more comprehensive. Many existing systems solve the problem of estimation of traffic mathematically, by using positions of different cars, average speed of cars passing, etc. as statistics. We try to solve the problem as humans do. When we, humans, look at a lot of cars on a road we know it’s heavy traffic. Similarly, with recent advances in computer vision and machine learning, we can train computers to do the same. We strive to develop a program that can look at an image of a street and automatically decide whether it is congested or not. We propose to implement a model that uses a convolutional neural network (CNN) to decide whether the given image is of a congested street or a free street. We further hope to deploy this model on a Raspberry Pi Mini Computer with an attached camera, which will be installed in the street corners. Thus, the Raspberry Pi will capture images of the street at brief intervals, use the CNN to estimate traffic density and then relay this information to a central server, where further processing can be done.