Artificial Intelligence: revolutionizing the world and saving wildlife

Artificial intelligence never ceases to surprise in the negative, but also in the positive, as thanks to this algorithm, which protects endangered wildlife

Artificial intelligence (AI) is rapidly transforming numerous sectors amid spirited debates and critiques about its application. Consider, for example, the world’s first restaurant fully managed by AI. But what if we told you that this cutting-edge technology could also protect wildlife from vehicle collisions?

This is being accomplished by the machine learning algorithm YOLO, an acronym for “You Only Look Once,” developed by the Institute of Mathematics and Computer Science at the University of São Paulo.

The YOLO algorithm has been implemented for real-time animal detection to prevent road accidents, thus protecting both drivers and endangered species. It operates in a remarkably straightforward manner.

By utilizing a database created by researchers, the algorithm can classify animals encountered in the wild near roadways. The system identifies these animals and alerts drivers in a similar fashion to how traffic congestions and similar hazards are communicated.

Drivers might receive a notification on their smartphone or vehicle’s onboard computer alerting them to the presence of an anteater on the road, to give just one example. There is no need to actually see the animal and brake suddenly.

“Our system uses cameras along the road coupled with a laptop and is innovative in this regard. Drivers also face significant risk in collisions with large animals, often having insufficient time to react to avoid them,” explained Professor Rodolfo Ipolito Meneguette.

An evaluation study, published in the journal Scientific Reports, has already been conducted in Brazil for detecting the country’s endangered wildlife. The results were impressive, with the model’s accuracy exceeding 80%.

How YOLO works

The YOLO dataset was crafted using 1,823 photographs of Brazilian mammals from the public domain, enabling the algorithm to identify different species.

“We have created a database of Brazilian species and trained a series of artificial vision models to detect them. According to estimates from this center, about 475 million animals are killed on Brazilian roads each year,” stated Gabriel Souto Ferrante, the lead author of the study.

Various versions were tested in the São Carlos Ecological Park using images and videos from camera traps. Surprisingly, older models proved more accurate than more recent ones.

The researchers now want to carry out new evaluations and tests through an application they developed which, unlike other very common navigation apps, will allow them to be updated by inserting new data.