They say the best part of a project is when everything comes together in the end and functions as expected. However we believe that…
...actually, they are right. But the second best part of a project is getting to show others what we have been working on. And that is exactly what we got the chance to do on the 9th of last December at the Széchenyi István University.
Since we had to travel from Zalaegerszeg, we had to miss the kick-off of the conference. We joined at a roundtable conversation about UAV laws, education, and its current issues within the European Union. Representatives of the Hungarian Drone Coalition and Légtér.hu answered questions and presented improvements to the current system for discussion. Precision agriculture also got a mention, as the newest emerging field involving UAVs.
Next up was a project developed at SZTAKI. This project, called Forerunner Drone, uses the same UAV platform as us to help emergency responders prevent accidents by giving them an autonomous eye in the sky that watches for upcoming traffic. They showed how their system was able to detect potential issues (like a driver heading into an upcoming intersection) and classify them by severity. It was even able to calculate whether the ambulance driver had to yield to prevent a collision.
They ran into the same issues we did in accessing the UAV's onboard camera systems, so it was quite helpful to discuss our solutions after the event.
Up next was our presentation.
As part of the GINOP programme, we spent the last one and a half years developing a system for automatic detection of foreign objects on the ZalaZONE testing tracks. This would enable the maintenance team to skip the highly automatable process of manually checking every module before testing activities could take place on them. The current method is both time-consuming and expensive, and also requires a keen eye to notice any issues. By taking out the human element, we can truly guarantee that a track is safe to use and free from any potential (if highly unlikely) obstructions, in addition to lowering cost.
Our system utilizes photos taken from the air using UAVs to automatically detect foreign objects and mark them on a map. This way track maintenance can get a complete overview of the current condition of a given module without leaving their office. Cleanup is still manual of course, but this still saves quite some time, not to mention pricey fuel, since our UAVs are fully electric.
We presented a custom machine learning model for object detection, and a web-based processing pipeline to visualise results. We were unfortunately unable to show a live demo due to time constraints, but we did present some earlier results and the basic use of the system.
After some quick Q&A, the main programming was over. All that remained was some networking and we headed back home, with yet another conference under our belts.