Computer vision and artificial intelligence are the core technologies behind what we are building at footpath.ai. As part of a cloud-based workflow of mapping sidewalks at scale, we have trained a visual semantic segmentation model for urban scene understanding specifically for understanding the complex environments on sidewalks.
Our visual semantic segmentation along with object detection models which are trained and run on NVIDIA GPU-powered AWS cloud services automatically detect pedestrian crossings and various types of street furniture at scale. Combined with the geo-reference information from each individual captured panoramic image and depth estimation, we infer the approximate location of the detected object/feature in the space. This allows us to produce a very detailed and connected sidewalk network that can be used for different applications.
In 2021, Google announced a few major updates to its map products including "detailed street maps" that contain information on where sidewalks, crosswalks, and pedestrian islands are. At footpath.ai, we go beyond sidewalks and crosswalks. We map a lot more details such as where accessibility ramps (a.k.a. curb cuts) are as well as detecting street furniture including trees, bike racks, benches, and garbage bins to name a few. This makes travelers understand, enjoy, and connect with the urban space a lot better.
Stay tuned for even more exciting updates and news over the next few months.