Who is more likely to lose a trail in the wilderness, a human or a UAV? If you guessed that humans are more likely to get lost then you would be right. In Switzerland there are over one thousand distress calls every year from hikers who have gotten lost or injured. Researchers at the University of Zurich have trained drones to identify hiking trails and follow them completely autonomously, like the reconnaissance drones in Prometheus.
The UAVs in this project were only equipped with forward facing color cameras to discern the trail in front of them. Using a supervised image classifier (with a library of over 20,000 trail photos) the system computes the main direction of the trail ahead compared to the viewing direction. The image classifier assigns one of three classifications to each image; either the trail is to the left of the viewing direction, straight ahead, or to the right of the viewing direction. Discerning trails from camera images is difficult for computers and humans, but not surprisingly computers do this better. Drones identify trails correctly 85% of the time whereas humans only identify trails correctly 82% of the time.
On a large trail network it can be difficult for a search and rescue team to check every fork and side trail. Emergency responders could use a fleet of UAVs to quickly explore a wilderness area and locate missing hikers.
Alessandro Giusti, Jérôme Guzzi, Dan C. Ciresan, Fang-Lin He, Juan P. Rodríguez, Flavio Fontana, Matthias Faessler, Christian Forster, Jürgen Schmidhuber, Gianni Di Caro, Davide Scaramuzza, and Luca M. Gambardella. A Machine Learning Approach to Visual Perception of Forest Trails for Mobile Robots. IEEE Robotics and Automation Letters. February 9, 2015.