A team of researchers at the University of São Paulo's Institute of Mathematics and Computer Sciences (ICMC-USP) in São Carlos, Brazil, is building a computer vision model that automatically detects wild animals on Brazilian roadways.

To accomplish this, the team created a database of Brazilian species and trained several computer vision models to detect them. To further avoid the risk of collision with larger animals on Brazilian roadways, specifically, where visibility might be poor, the system also relies on roadside cameras paired with a portable computer.

General architecture proposed by Bochkovskiy et al.37, focused on object detectors based on one-stage and two-stage convolutional neural networks. (Adapted image). Source: Scientific Reports (2024). DOI: 10.1038/s41598-024-52054-yGeneral architecture proposed by Bochkovskiy et al.37, focused on object detectors based on one-stage and two-stage convolutional neural networks. (Adapted image). Source: Scientific Reports (2024). DOI: 10.1038/s41598-024-52054-y

To build the database of Brazilian mammals that are most likely to be hit on roadways, the team downloaded 1,823 images available in the public domain. These were edited to remove "noise" that threatened to hinder species identification.

Different iterations of the computer vision algorithm, You Only Look Once (YOLO), which is commonly used for the real-time detection of objects, including animals, were then tested.

The videos of animals captured by the researchers were subsequently used to test the system, with the researchers suggesting that future versions of the database will include images of animals captured by forest camera traps and roadside cameras.

During testing, older iterations of YOLO were deemed better at detecting animals.

“The models correctly detected the species in 80% of images taken during the day with the animal appearing clearly," the researchers said.

New images will be added to the database, and the researchers will partner with toll road operators and city governments to test the animal detection technology in real-world situations.

The research is detailed in the article, "Evaluating YOLO architectures for detecting road killed endangered Brazilian animals," which appears in the journal Scientific Reports.

To contact the author of this article, email mdonlon@globalspec.com