David Recchia is a 3rd year student in the Geomatics Engineering program at the Lassonde School of Engineering at York University. Having extensive experience in land surveying, this summer he chose to explore other applications within the realm of Geomatics. His curiosity has led him to Dr. Sohn’s GeoICT lab. While assisting his lab colleagues in their research initiatives, David is also working towards developing a road auditing pipeline that takes advantage of the recent advancements in computer vision, deep learning and photogrammetry. By the end of the summer, he is hoping to a build a program that automatically detects/locates objects specified by the user along a roadway using Google Street View images. This research initiative is important for two reasons—first, to help facilitate street audits in order to detect road assets/features that impact cyclist safety. Secondly, to demonstrate the applicability of using images to conduct cost-effective street audits at the city scale.
Road Condition Mapping for Bicycle Safety Using Deep Neural Network and Google Street View
Since the adoption of the Vision Zero road safety plan two years ago, little progress has been seen on the City of Toronto’s roadways in terms of traffic incidents. Currently, the number of pedestrian fatalities is at 17 and there have been three cyclist deaths this summer so far. Toronto is at a critical point in terms of road safety—it needs to address the lack of infrastructure that exists in order to protect cyclists and prevent future incidents. This, however, is a daunting and costly task. Roadways used by cyclists need to be assessed for any potential hazards or concerns that are present—such as catch basins, potholes, and narrow lanes—which require laborious activities. With the unprecedented amount of coverage that Google Street View has worldwide, this free database can be used to facilitate a multitude of new investigations and inventory collection for cities. In fact, a growing number of investigators have used Google’s aerial and street view images to create an inventory of trees, assess the accessibility of bus stations and even determine the socioeconomic demographic of neighbourhoods. Given this, an opportunity is presented to build a program that is able to conduct road surveys from the confines of someone’s office. It is, therefore, the intention of this endeavour to develop a user-friendly program that automatically detects and geolocates user-specified objects from a given street. This is made possible through the recent advancements in deep neural networks and fine-grained detection. For the purpose of this task, cyclist safety will be the main focal point. In doing so, images of specific streets in Toronto will be collected using several of Google’s APIs and used to automatically detect and locate catch basins that lie within cyclist pathways.