© The Italian Coastguard / Massimo Sestini

23 October 2019

2.5 days, 55 AI experts and one goal: saving lives in the Mediterranean.

This summer, Deep Berlin hosted the first  AI  for  Good  Hackathon, a collaborative event for the development of hardware and software for a good cause. This year's mission: To help sea rescues locating boats affected in the Mediterranean.

Still many people are dying in the open sea, in the last five years it has been 16,000 and the numbers are rising. Boats can be identified faster and lives can be saved, with the use of deep learning software for image analysis in drones. "Manually searching ships on the horizon using binoculars is extremely time-consuming and requires a high level of concentration. Corresponding radar systems on the ships do not have the accuracy to detect small boats at great distances. Unfortunately, a search by aircraft is not possible at all times. The use of artificial intelligence in automated flying drones is an efficient way to detect boats and distinguish them from waves, cloud shadows, etc.," explains Dr. Mustapha Al Helwi, initiator and partner of Deep Berlin.

Image processing software for drones is already used in sea rescue, but classic algorithms are reaching their limits. This led to the idea of using the AI network of Deep Berlin and finding competitors for the advancement of the software as part of a hackathon. More than 80 applications were received within three weeks, although there was no economic incentive. "It is remarkable that 55 AI experts, some of whom had travelled to Berlin from Stuttgart, Munich and Augsburg for this event, sat in a room in front of the computer in great weather in the summer and spent their entire weekend working for a good purpose," Al Helwi said.

Currently, the software is in a beta test with selected smaller rescue ships. Beyond the hackathon, some participants even continue to work on the algorithms to support the project in the long term.

"A manual creation of algorithms that can represent the same recognition performance would take years and is not desirable due to the severe situation," Al Helwi explains once again. However, as in most cases, it is not possible to rely solely on artificial intelligence. All drone images have to be seen manually. However, through selection and separation artificial intelligence offers a great relief and increases the efficiency of the sea rescue.

Due to the great success of the first hackathon, the organizers are confident of hosting another Hackathon in the coming year. Then the motto will be again: Do good with AI!