Laszlo Talas’ research focuses on computational approaches to visual perception, including animal, human and machine vision. With a background in zoology and experimental psychology, he is particularly passionate about how visual scenes can be “understood” using computers and what comparisons can be drawn with biological visual systems.
Understanding vision can help us to generate a positive impact on the world, for example, automatic disease detection systems to improve animal welfare, providing a better museum experience for visitors, or raising awareness of how the colours of animals work.
Laszlo, congratulations again on becoming one of our first cohort of University Enterprise Fellows! Tell us about your career to date.
Thank you! My career track is fairly simple: I came to Bristol as an undergraduate in 2008 and I’ve been here ever since, climbing the academic ladder.
I did a degree in Psychology and Zoology, and then a PhD in Biological Sciences. Well, that’s what is written on the certificate! Actually, my PhD was in the cultural evolution of military camouflage, so it was fairly different from the other Biology PhDs going on around me. It was more along the lines of anthropology really. After my PhD, I continued working on camouflage as a postdoctoral research associate. The project was focusing on using artificial intelligence techniques to explore the optimal camouflage in any given environment.
After that I decided to translate my research into the veterinary world. That was quite an ambitious move. To make it even weirder, a colleague and I decided to apply for an EPSRC Fellowship jointly. Two people applying for a single fellowship with no veterinary background! Fortunately, EPSRC bought it. They funded us to explore how thermal cameras can be used to track respiratory disease in cattle. And that’s how I ended up in the Vet School, as an ESPRC Fellow and now a Lecturer in Animal Sensing and Biometrics.
You’ve worked across multiple disciplines throughout your career, and have “discipline-hopped” more than once. Was that something you set out to do from the start?
I aspire to being active across multiple fields of research rather than being a specialist in just one. I’m continuously looking for new people to collaborate with. I find Bristol to be amazing in this aspect – it provides great platforms to collaborate across disciplines, and that I really, really value.
I did a joint degree and a very interdisciplinary PhD, so I grew up in an environment where I was encouraged to talk to people in different departments. At the very beginning of my PhD, my supervisor invited me to join the Bristol Vision Institute, which helped me grow my network of academics. I became a postgraduate rep for the management team, which helped me to meet people from other schools. I also went to lectures across the University, for example Engineering undergraduate lectures, just to learn about image processing. I didn’t understand 90% of it, but everyone was very welcoming! People around me pushed me to be interdisciplinary and after a while it becomes second nature.
It’s not just other academics either. I like to work with the people in the mechanical and technical workshops. There are tools around at Bristol that are open for everybody, but a lot of people just don’t know about them. So, I tell my PhD students: “OK you want to install a CCTV camera to monitor chickens? Fine, there are two ways you can do that. You can buy one from the shop. But we can also build a camera. It’s cheaper. We can programme it the way we want it. We need a custom-made box for it? We’ll build one. We’ll order acrylic sheet and go to laser cut it ourselves. It’s not just cheaper, but you will get exactly what you want.” All this infrastructure is there, the University is an enormous resource for its researchers.
Tell us about what you want to achieve in your fellowship, which concerns automated behaviour monitoring in horses.
There is a personal dimension to this. While I work in a vet school, I’m not a vet. However, my late grandfather was a vet and my father is a vet. I was never compelled to become a vet myself, but I’ve grown up in an environment with lots of animals and I was inspired to work with them in some way.
After collaborating with my father on a paper, I learned that there’s a lot of interest in thermal photography in veterinary applications, but also a lot of bad practice. Sometimes people take a single thermal picture of an animal and then draw conclusions and diagnoses based on that. It occurred to me that we could buy many small, cheap thermal cameras, connect them with Raspberry Pi computers, and we could take millions of images of animals fairly easily. We can just install them on a farm and watch the animals continuously as they go about their day and monitor their health.
My research collaborator John and I got really interested in developing this as a means to assess lameness in horses, so we visited the Vet School to talk to veterinary academics. To secure funding, they recommended we focus on livestock, of national interest – so we turned from horses to cattle. We were initially thinking to look at lameness, but the vets recommended we focus on respiratory disease. We didn’t mind at all – we were not that married to the idea of a particular animal species or condition, we just wanted to use the tech! Whether it was a stable or a pig pen or a chicken barn, it didn’t matter to me too much. It was good advice, and we managed to get an EPSRC Innovation Fellowship to take the cattle application forward.
With the benefit of all that research, the focus of the fellowship is back to the horses. Recently, we’ve been working closely with vets at Newmarket. We reached out to Newmarket because it’s a world-renowned place for horses, both in numbers and value. Our main contact there has been extremely supportive, and it turned out he is also a Bristol graduate. We installed thermal cameras in horse stalls of a thoroughbred bloodstock and took videos of mares and foals. The focus is to automatically detect and classify behaviours of horses.
What sort of behaviours are you looking for – what are you hoping to see?
To start with, simple stuff: is the horse standing or lying down? Is it active? Then focus on more detailed behaviour, for example to detect stress. Can we detect the horse chewing on the gate? Kicking the stall? The system could let us detect signs of illness, or anxiety – a whole range of different problems.
But there’s also value in simply mapping the day of a horse, because we do not know exactly what any one horse should be doing. A textbook might say that a horse is supposed to sleep for so many hours, eat for so many hours, but there are loads of individual differences. One appealing prospect is to track horses continuously for several months and see if there are any deviations in their behaviour. All of this can build up a bigger picture of what is normal so we can detect when something is wrong.
Horses are highly valuable. The horses we work with have some of the best care in the world, they are surrounded by specialists all the time. But things still go wrong, they still have problems. If we can detect those problems earlier and solve them faster, it could be a big win for the owners and the stables as well as the vets.
You can look at automatic monitoring technologies in two ways:
The first thing we can do is to automate something that a human can do, but it’s impractical, for example it takes an unfeasible amount of time. Let’s say that you have some sort of diagnostic, like how droopy a foal’s ears are. You can, technically, ask somebody to stand there 24/7 and stare at the animal’s ears, but that’s just never going to happen, it’s not practical. If you have a system that takes the mundane work away but provides the data, vets can use the data but meanwhile focus on more productive things.
The second thing we can ask is whether technology can detect signals that the human would struggle to spot. Can we detect disease earlier? Can we do something that’s beyond the human capability? That’s very appealing – can the vets get a red-light warning on the wall to prompt them that an animal needs attention before manual methods would indicate that anything is wrong?
It’s important to note that the applications aren’t limited to racehorses! A place like Newmarket is a great environment to start with; they have lots of animals, all the infrastructure like electricity and internet, which makes it easy to collect data. However, my goal was always to focus on low-cost technologies that can ultimately be used almost anywhere and work towards products that are affordable globally. That’s why I’m a big fan of putting technology together in a “do-it-yourself” fashion. Cheaper and more adaptable.
What’s the key next step?
Academia is what I am familiar with. I do research, but I think of the research as the tip of the iceberg, and I have no experience of the part that is under the water – product development, advertising, setting up companies and so on. As an academic I could develop a great camera system, develop an automatic behaviour classification system, publish the paper and move on with life. I thought, well, maybe I can do more than that: oversee the entire process from having an idea, carrying out research to actually designing, manufacturing and selling a product. I have no experience in the latter part, but I am keen to learn more. In short, that’s why I applied for the fellowship.
I would like to go and talk to people in the equine industry to find out what their specific needs are. The list includes equine vets, stable owners, and individual owners. Rather than second guessing what the world needs, I want to use this opportunity to seek out what the actual problems are and whether I can contribute to solving them.
One more question, Laszlo: you call your technology AutoNonius. What is the significance of the name?
Ohh that! It’s just a word play really. The Nonius (or Nóniusz) is a Hungarian breed of horse (Hungary being the country I originate from). I liked the pun with “auto”, although of course what I’m doing is automatic and not autonomous! It might have to change down the line, but it is an alright working title.
Laszlo Talas is Lecturer in Animal Sensing and Biometrics at Bristol Veterinary School.