Computer vision algorithms are helping scientists automatically identify rare Saimaa ringed seals through patterns on the seal’s fur. The aim of the study at Lappeenranta University of Technology (LUT) in Finland is to get a biometric passport for each individual Saimaa ringed seal, to track their movements without physically attaching transmitters.
The Saimaa ringed seal is an endangered species and, until now, any identification has been done manually by placing transmitters in the fur of the seals. Manual identification is nevertheless time consuming and difficult. Furthermore, the transmitters can interfere with the animal's activities to some degree.
‘Now the purpose is to help the identification of the seals in such a way that the fur can be found in the pictures, so that the fur of each seal becomes its biometric passport,’ commented Heikki Kälviäinen, Professor of computer science and engineering at the Laboratory of Machine Vision and Pattern Recognition at LUT. ‘The Saimaa ringed seal can be in any position, which means that we need to achieve identification that is independent of position, using a computer's machine vision application. First we need to locate the seal in a picture. After that we need to classify, or identify, which individual out of about 300 Saimaa ringed seals is involved.’
According to Professor Kälviäinen, initial research results indicate that automatic identification is fully feasible. Automating the identification is first tested in a testing environment from several photographs of about 300 seals. Researchers have a total of more than 1,000 such pictures. Once the automated system is ready, final identification is made on the basis of game and wild animal camera pictures.
‘In nature the challenge is that it can be dark, and the background of the picture can blend in to the seal. A seal can also be in a position in which it is hard to identify,’ added Kälviäinen.
Automation would make it easier for biologists studying the Saimaa ringed seal to know if an individual animal appearing in an area is new or, for instance, what an individual seal that has died had done before dying. When individuals are identified electronically, it will also be possible to confirm the mating of the seals more easily than before.
‘At the moment we are engaging in basic research, but ultimately the goal is to find a computer-based automated solution for the identification of the seals that is as flexible as possible,’ Kälviäinen said.
Further information:
