Challenge

Kauê knows the important role sidewalks play in pedestrian safety, and also knows that very few places have detailed maps of sidewalks. Major cities across the world lack pedestrian data that accurately describe conditions on the ground. Route optimization, infrastructure planning, and incident reporting are a few ways cities can improve safety and accessibility with better data. Kauê’s interest lies with improving data to make sidewalks safe and accessible for everyone.

While aerial imagery is useful for adding basic geometry and validating network connectivity among a pedestrian network, features such as tall buildings, trees, and shadows block mappers from seeing all the details. Curb cuts, signs, and signal markers can only be seen from the ground. The more detail a pedestrian dataset has, the more ways it can be used, and the more options people have when traversing a city.

Solution

Building detailed data is a massive amount of work, especially in Curitiba, Brazil where it is common for sidewalk conditions to vary from house to house. Realizing the massive undertaking before him, Kauê was determined to find the most efficient way to build better data. He quickly realized how useful Mapillary was after experimenting with a custom mobile mapping system mounted on top of a friend's VW Beetle. Kauê began capturing streets in Curitiba where Mapillary imagery was missing before moving on to data extraction.

At the beginning of his PhD he created OSM SideWalkreator, a QGIS plugin guiding mappers by creating basic sidewalk geometry, curbs, and crossings that can be exported directly into JOSM. Kauê began extracting sidewalk geometries generated by his plugin, and solicited editing assistance from his local YouthMappers chapter, Mapeadores Livres UFPR. Students used Mapillary imagery to edit OSM using Pic4Review , a platform that displays street level images for adding features best seen from the ground. Mappers tagged curbs with “raised vs lowered” while adding surface type and condition to sidewalks. They also tagged crossings with marking type and whether they had traffic signals. These details drastically improved results to routing algorithms used in his research.

To further support his work Kauê built OpenSideWalkMap (OSWM), a platform that represents OSM sidewalk data on a Map while offering other tools for pedestrian data management. This tool is meant to improve routing for people using different modes of transportation. Maybe a mother pushing a stroller prefers a route that is all asphalt and concrete, or a person riding a scooter may want to avoid cobblestones. OSWM can optimize routes for these scenarios but only if detailed tags exist.

Inspired by these initial results, Kauê began researching how machine learning could augment the manual process. He is now creating a Framework called Deep Pavements, which aims to identify pavements and surface types using AI workflows, the main module of the project is available at deep pavements project.

As Kauê continues improving his tools and workflow, he plans to scale his work in other cities throughout the world. He is spending most of his time nowadays improving automated sidewalk generation with attribute extraction from street-level imagery, and is interested in connecting with other researchers. If you are working in this space, like to capture street-level imagery, or are passionate about improving pedestrian data in your city, Kauê would like to hear from you. The best place to find him is on Linkedin or Github.

/Chad