Ingo Bartussek - Fotolia

How drones and machine learning can prevent crocodile attacks

Cloud-powered drones equipped with machine learning algorithms will soon be able to detect crocodiles lurking in the waters of Queensland, Australia

Cloud and machine learning are being used to interpret drone-collected images that will be used to scan for crocodiles in a bid to protect swimmers and tourists in Queensland, Australia.

At the World of Drones Congress in Brisbane, a demonstration of the CrocSpotter technology successfully identified crocodiles in the Mowbray river near Cairns in support of the Queensland government’s CrocWise initiative.

Intended for use on beaches and waterways from Mission Beach to Port Douglas at its inception, CrocSpotter will be rolled out over the next four weeks as part of a three to four-month trial, during which its machine learning algorithm will be trained and enhanced.

Ben Trollope, CEO of the Ripper Group, a Westpac-sponsored operation that uses drones for safety and rescue missions, said there have been 97 crocodile incidents in the past 100 years.

He stressed that the information collected through the operation would help inform agencies about the location of crocodiles and steer them towards the most appropriate course of action.

But although CrocSpotter is intended to help prevent crocodile attacks and encourage tourism, it also has an important conservation role. “If we know a shark or a crocodile is there, we can track it and provide real-time situational awareness,” said Trollope.

Underpinned by the Amazon Web Services (AWS) cloud, CrocSpotter uses the Ai Spotter algorithm initially developed by the University of Technology of Sydney (UTS) and the Ripper Group to identify sharks.

With CrocSpotter, Ai Spotter is deployed on Westpac Little Ripper drones that transmit live video streams to ground pilots. The algorithm analyses the footage and singles out potential threats via a flashing red box around a crocodile.

The images collected by the drones are also uploaded to the Amazon S3 cloud storage service, where they can be used for machine learning, and to enhance the crocodile recognition algorithm.

Read more about cloud computing in Australia

Ben Thurgood, head of solution architecture at AWS, said CrocSpotter has been proven to be 93% accurate, compared to human spotters, who have been achieving a 19% strike rate at best.

Also, it can achieve less than one second latency between a crocodile being identified and a lifesaver being alerted, while tapping the scale and reach of the cloud to stream live images anywhere.

The UTS researchers who designed the algorithm said the use of cloud also makes it possible to detect crocodiles in a range of environments, including murky and muddy waters in wetlands and in the open ocean.

According to the Business characteristics survey conducted by the Australian Bureau of Statistics, the appetite for public cloud services in Australia continues to grow, with 42% of businesses reporting the use of cloud computing, compared with 31% in 2015-16.

Read more on Artificial intelligence, automation and robotics

Data Center
Data Management