Computer vision system weighs pigs without a scale

Glacier FarmMedia – The three metal-framed stations in the pig barn conjure the image of a starting gate at the Kentucky Derby. A pen opens. For a minute, the air is full of galloping hoofs and excited grunts.

Twenty pigs run through three short aisles equipped with lights and cameras, and pool in the pen on the opposite side.

Franςois (Frank) Decarie, one of the research partners developing a new low-labour weighing method for livestock, leans over a table beside the stations. He hits play, and a recording of the rush of pigs pops up across three monitors.

Footage from the aisle cameras feeds into computer models trained to identify individual pigs, he tells attendees of his demonstration at the Bruce D. Campbell Farm and Food Discovery Centre near Glenlea, Man.

The models use that data to draw an outline through the animals and draw shapes around key anatomical features like hams, shoulders, spine and first rib. The system then combines measurements to estimate weight.

All said, it took handlers longer to get all the pigs in the same pen than it did to weigh them.

Francois (Frank) Decarie demonstrates the computer system behind the Porcus Optimus scale project. Photo: Geralyn Wichers

Decarie is a veteran of the Royal Canadian Air Force and worked as a communications electronics engineer. The research project, which he and his research partner, Charles Grant, call Porcus Optimus, is part of his doctoral thesis in electrical and computer engineering.

The project is backed by Moccus Maximus, a company Grant owns with an investor.

Computer vision

The system has undergone many run-throughs and has learned which measurements best predict animal size. It weights the data accordingly, demonstration attendees heard.

It’s not perfect. In one case, Decarie noted, the computer drew an outline around two pigs instead of finding individuals. When asked about the technology’s accuracy, he said they expected less than a plus or minus 1.8-kilogram deviation “with a high degree of confidence.”

The two researchers are still validating Porcus Optimus at the Glenlea location.

The stations are still fitted with conventional scales. The computer uses the camera to judge when the pig is in the right place and lines that up with fluctuating input from the scale.

A close-up view of what the computer “sees” as the pigs run through the scaling stations. Photo: Geralyn Wichers

“When they gallop through, we can still be accurate,” said Grant, a senior instructor in the University of Manitoba’s agribusiness department. “The pigs love to go as a herd through the stations.”

The scale is not required anymore, he added, but it’s there to feed more data into the system.

Getting smart

The computer vision system is a speedy method to weigh and count a pen or truckload of pigs. However, Decarie and Grant suggest it could do a lot more. The data could be fed into an optimizer, which could be used to predict ideal shipping times or suggest tweaks to feed regimens.

The pigs in the Glenlea barn have RFID tags, which allows the system to assign weights to specific pigs and could allow for individualized growth curves. However, since Decarie and Grant expect many producers won’t want to tag pigs (the tags aren’t compliant with PigTrace, the pork industry’s standard traceability system), they are also running trials without the use of individually tracked pigs.

“That way we’ll be able to actually measure if there’s a value of having tags at all,” Decarie said. “Eventually, we’d like to be able to ID the pigs with just a camera, and we’re exploring other avenues.”

Decarie and Grant custom-built the weigh station for their space at Glenlea. The idea is that, once commercialized, each producer will tailor the Porcus Optimus to their needs. Grant and Decarie envision stations in a barn’s loadout room or on an arm that can be swung over an alley.

They expect to sell the computer vision system as a service rather than a finished product, Grant said, and will continue research to continuously improve the system.

The scaling stations would be in the barn and the data could go onto the company’s server, which could generate reports for the producer or send notifications.

Decarie and Grant plan to present some of their research at the Banff Pork Seminar in January and begin commercialization some time in 2025.