ANIMAL research could be transformed by a Big Brother- type system that can monitor creatures 24 hours a day. The technology would reveal the effects on behaviour of drugs or genetic variations with unprecedented detail. It could also improve welfare, alerting staff when lab animals show signs of stress or disease.
鈥淚t allows you to look at subtle changes in behaviour that we know are occurring but cannot study because of the labour-intensive nature of long-term monitoring,鈥� says Tony Yaksh of the University of California, San Diego, a pain researcher who hopes to use the new system.
The 鈥淪mart Vivarium鈥� is being developed by a team led by Serge Belongie, also at the University of California, San Diego. The prototype is a normal camcorder hooked up to a desktop computer. The final devices will resemble PDAs, consisting of a small computer with a built-in camera that can be attached to the side of a cage. This will process video footage and send the results wirelessly to a central computer. Hundreds of cages could be hooked up to the system, so the behaviour and well-being of thousands of animals could be monitored at the same time.
Advertisement
The prototype can track several mice at a time and determine whether they are walking, sitting or stretching. It can even count how many steps each mouse takes. But the real breakthrough is that the system can keep track of individuals even when they cross in front of each other. This might seem simple, but tracking similar-looking objects when they overlap is a major challenge in computer vision.
鈥淚t鈥檚 hard for people to do this. It鈥檚 even harder for a computer,鈥� says computer vision expert Tucker Balch at the College of Computing at Georgia Tech in Atlanta, Georgia. His team has created a system for tracking ants (快猫短视频, 9 June 2001, p 21), but it can get confused when their paths cross. 鈥淲e are working on several solutions of our own,鈥� Balch says.
Belongie鈥檚 software assigns an ellipse to each mouse, according to how many pixels it occupies in each frame. When one mouse walks past another, the program assumes the lower mouse must be in the foreground. Other indicators such as fur patterns and colour help ensure the program gets it right. In tests, the system has successfully kept track of up to three mice in a cage (see right). Belongie is now collecting video footage of all relevant behaviours, so the program can be trained to recognise movement patterns characteristic of, say, stress or fear.
The system is unique in that it does not require special cages to be built, Belongie says. What鈥檚 more, it means animals do not need to be taken out of their living quarters to study their behaviour, which is often the way experiments are done. This is seldom desirable, as it stresses the animals and can confuse results. 鈥淚f an animal has a panic attack every time you place it in a water maze, you might conclude that it is learning-deficient,鈥� says neurobiologist Laurence Tecott of the University of California, San Francisco.
Tecott has developed cages with motion-sensitive platforms for tracking animals. It can monitor where mice are and also detect if they are eating or drinking. The system could never detect the fine behavioural details that Belongie plans to monitor, but it does prove the value of tracking mice continuously. Tecott says his unpublished analysis of 16 mouse strains has revealed 鈥渞emarkable differences鈥� in their lifestyles.
Belongie鈥檚 system could also come in handy for monitoring livestock on farms. And his approach could help researchers trying to develop CCTV-based systems, where keeping track of people as they overlap is also a major challenge. Belongie hasn鈥檛 solved the whole problem, says computer vision expert Michael Isard of Microsoft Research in Silicon Valley, but his system will probably solve some people鈥檚 problems.