The AI eye-in-the-sky turns to croc detection
In a world first, a drone fitted with the revolutionary CrocSpotter artificial intelligence (AI) algorithm has streamed a live video seeking and identifying crocodiles at an accuracy of 93 per cent and a latency of less than one second.
CrocSpotter AI can detect crocodiles with 93 per cent accuracy. Image Joshua Cotten on unsplash
CrocSpotter AI is the latest algorithm developed by the University of Technology Sydney and industry partner Westpac Little Ripper.
Ultra-low latency streaming is achieved via the Amazon Web Services (AWS) cloud, and this quicker delivery of vital information to search and rescue unmanned aerial vehicles (UAVs) could mean the difference between life and death.
Teams at two locations achieved the successful live video stream: the drone team flew over a known crocodile habitat at Mowbray River in North Queensland, while the other team received the stream 1690 kilometres away at the World of Drones Congress at the Brisbane Convention Centre on 26 September.
The drone carries CrocSpotter AI as part of its payload with the on-board video. As video streams in real-time from the drone to the pilot on the ground, the algorithm operates by “washing” the video, and alerts the pilot to a possible threat below.
The threat is highlighted immediately by a flashing red box around the detected animal, drawing the pilot’s attention to that part of the screen.
This is the first time this sort of animal detection drone technology has been deployed via a high-quality video stream at ultra-low latency, with the AI producing 93 per cent accuracy to detect crocodiles.
Dr Nabin Sharma
The algorithm has a 93 per cent accuracy of identification, compared to the naked human eye, which is between 16-19 per cent.
Professor Michael Blumenstein and Dr Nabin Sharma from the UTS Faculty of Engineering and IT first worked with Westpac Little Ripper to develop SharkSpotter, the AI-powered drone-based technology for protecting beaches and keeping swimmers and wildlife safe.
Continuing this effective collaboration, they are now developing cutting-edge AI software deployed via drone on the Amazon Web Services (AWS) cloud for detecting crocodiles from Mission Beach to Port Douglas.
The Queensland Government is keen to see the roll-out of the technology to prevent crocodile attacks, and to develop a better understanding of crocodile populations from a conservation perspective.
They are using an AI technique called deep neural networks to detect the crocodiles from moving drones, via a smart camera, and have deployed the technology in the cloud.
The live stream from the CrocSpotter to the World of Drones Congress in Brisbane
Professor Blumenstein said the speed of the cloud-based AI can spot crocodiles in real-time, which is a world-first and a technological breakthrough given the very low latency.
Dr Sharma said the technology enables crocodile detection in complex environments, including murky and muddy waters in both wetlands and the open ocean.
“This is the first time this sort of animal detection drone technology has been deployed via a high-quality video stream at ultra-low latency with the AI producing 93 per cent accuracy to detect crocodiles,” he said.
Ben Trollope, CEO of The Ripper Group, said the tripartite cooperation means a latency of just one second for first responders.
“Running the CrocSpotter AI algorithm via AWS reduces the latency from 10-30 seconds currently achieved by the drones’ on board video,” he said.
