Today drones can maneuver indoors, around difficult to reach nooks, bends and even dense forests too. What’s more? Well, recently drones again hit the headlines, owing to their new ability to help people, hikers lost in woods. In their paper published in the journal Nature Machine Intelligence, researchers, David Schedl, Indrajit Kurmi and Oliver Bimber, from Johannes Kepler University, share how artificial intelligence to improve thermal imaging camera searches for people lost in woods.
When hikers, trekkers or commoners are lost in woods, rescue team rely on binoculars, and thermal imagers installed on camera and in chopper sensors, to find the missing. Generally, the thermal imaging devices highlight differences in body temperature of people on the ground versus their surroundings, making them easier to spot. However, the problem with these forms of devices is that they fail to work efficiently due to dense vegetation covering the soil and preventing wide or long-range view searches. Additionally, in the case of thermal imaging, heat signage from external factors like the sun, ambient heat from trees, fauna and other humans’ body heat also interfere with the search process. Further, these existing tools used in such search and rescue operations are not sufficient nor speed up the mission.
The trio team of researchers hence focused on surmounting these problems by using a deep learning application to improve the images that are made. They engineered drones which can be particularly adept at recognizing humans from everything that surrounds them. They aimed to show how automated person detection under occlusion conditions can be notably improved by combining multi-perspective images before classification.
They employed image integration by airborne optical sectioning (AOS). AOS is a synthetic aperture imaging technique that uses camera drones to capture unstructured thermal light fields. In simpler terms, they used AI application to process multiple images of a specific area. This is interesting because comparing and processing of data from different cameras allow several thermal imagers to work as one large telescope.
After processing the artificial intelligence images, the final terrain images produced have a higher depth of field, i.e. the footage shows how the tops of the trees appear blurry, while the outlines of people on the ground were more recognizable.
For training the artificial intelligence system, the researchers had to create their own image database. They used drones to photograph volunteers on the ground in a variety of positions. The system test results showed that the accuracy of this solution ranged from 87%-95% when compared to 25% of traditional thermal imaging images. Also, using AOS helped researchers achieve a precision and recall of 96% and 93%, respectively.
“In the future, rescuing lost, ill or injured persons will increasingly be carried out by autonomous drones. However, discovering humans in densely forested terrain is challenging because of occlusion, and robust detection mechanisms are required. We show that automated person detection under occlusion conditions can be notably improved by combining multi-perspective images before classification,” say the researchers. They also revealed that their system is ready for use by search and rescue crews and could also be used by law enforcement, the military, or wildlife management teams. Moreover, these findings can lay the foundation for effective future search-and-rescue technologies that can be applied in combination with autonomous or human-crewed aircraft.