Robotics engineer Stephen Roberts was taking his lunch at Somerville College at the University of Oxford, in England, when the conversation turned to chicken. It wasn’t the food, though. His dining companion was animal welfare specialist Marian Dawkins, and she thought that the pattern-recognition technology Roberts was explaining might help identify misbehaving hens. Laying hens aren’t ordinarily antisocial, but under the stress of incarceration, they are known to peck one another, sometimes to death. It usually starts with a couple of chickens and can spread quickly. Once a bird starts to peck, others follow suit. For poultry farmers, the behavior is costly and difficult to deter. Many farmers resort to clipping the chickens’ beaks, but some countries ban the practice on humanitarian grounds. Environmental adjustments, such as dimming the lights or improving foraging material, can prevent attacks, but those work only if farmers know which hens need the changes, and each adjustment has its costs.
The key to detecting unruly hens might be in observing how they run, Roberts says: “Movement patterns [at an early age] are a proxy for behavior later on.”
Dawkins hoped that by recording the shuffle and flow of chicken flocks over time, Roberts might train his software to distinguish between healthy flocks and stressed-out, violent ones before the feathers flew. Previously, Roberts studied human crowd movement using a machine-vision system based on optical flow. Optical flow is a measure of the pixel-to-pixel changes between subsequent frames of a video. The principle is the same for humans and chickens: Each image is slightly different from the last, with humans or chickens moving relative to one another within the video’s frame, and each kind of crowd has its own signature flow pattern.
To show that the method could apply to hens, Dawkins, Roberts, and their colleagues did experiments in which some statistics from the optical-flow pattern correlated well with an established but much more arduous method: watching how individual chickens walk. Farmers can hire trained experts to evaluate the health of flocks by scoring each chicken’s gait, but those experts can’t constantly monitor a flock, says animal behaviorist Harry Blokhuis of the Swedish University of Agricultural Sciences, in Uppsala.
“It was an engineering challenge,” Roberts recalls. “Could we mimic the laborious labeling with a computer?”
Roberts turned to hidden Markov chains, a kind of mathematical model that uses noise in a data set to estimate the underlying “normal” signal and then identify when the signal is going awry. Then the team applied it to footage of more than 300 000 commercial free-range chickens. They found telltale behavioral disturbances in the flocks that had the most feather damage. Zoologists who saw the results “were certainly really surprised that we could forecast so well the prevalence of feather pecking,” Roberts says.
What isn’t clear yet is whether the method will provide an early enough warning of future problems, says animal behavior and welfare researcher Bas Rodenburg of Wageningen University, in the Netherlands. Fatalities in flocks seem to result from the tension between the way chickens are raised and some of their basic instincts. Chicks hatched into today’s system are likely to be raised in large numbers, in close quarters, and with very little training in how to be hens, he says. Since they can eat their fill with little effort, their age-old instinct to forage and peck for food all day doesn’t have an outlet. “It’s really important to stimulate healthy foraging early on,” Rodenburg warns, because birds who can’t forage the normal way—by pecking the ground in search of food—are likely to peck at other things instead, including their neighbors.
In recognition of these problems, in 2012 the European Union will outlaw the smallest poultry cages in favor of cages with more room for chickens to wander and express their foraging instincts. But Rodenburg notes that the trend toward free-ranging flocks may enable the most deranged hens to harm more of their coop mates. Dawkins and Roberts, meanwhile, are making larger-scale versions of their system and aim to share future detectors with farmers and animal welfare agencies for real-time monitoring of other variables and other species. “It doesn’t replace observant farmers,” says animal scientist Blokhuis, “but it could focus the farmer’s attention.”