From pixels to breeding values: can cameras help breed better hens?

Automated tracking technologies make it possible to perform large-scale genetic analyses of behavioural traits in livestock. However, these technologies bring their own challenges, for example identity switches and false positive or negative detections, which could result in a systematic misassignment of phenotypes. “While some studies have successfully estimated genetic parameters for behavioural traits in pigs using computer vision data, there’s a greater technical challenge when trying to track individual laying hens, as pigs are typically housed in smaller groups and are larger in size than poultry.” says Tzayhri Osorio Gallardo, first author of the study.

However, a potential solution to identity switches is the use of computer-readable markers, which are implemented in the publicly available software library called “ArUco”. To evaluate the suitability of ArUco-based tracking of individual laying hens for genetic analysis of behavioural traits, Tzayhri and her colleagues used cameras to continuously record their activity in the litter area. This allowed them to track where the animals were, how long they stayed, and how much they moved. From these videos, they extracted the position (x and y coordinates) of each hen in every frame. Consequently, they used this information to derive three behavioural traits: detected or not, minutes detected, and average walking speed. These traits were tested as potential proxies for activity in the litter area, a location which has been associated with behaviours that can indicate either positive or negative aspects of welfare. Tzayhri: “The estimation of significant genetic variance and of breeding values would demonstrate the existence of genetic effects on traits related to litter area activity, which could potentially be included in a breeding program.”

On average, individual hens were successfully detected 37% of the total observation time, with the proportion of time detected varying substantially among individuals. However, this does not mean that the hens were absent for the remaining time: in many cases, they were still within the litter area, but their markers were temporarily obscured or not recognised due to technical limitations, for example occlusion or camera position. Despite this imperfection, the high number of repeated observations per hen allowed the researchers to capture consistent differences between individuals. Tzayhri: “We achieved an average of over 5,000 detections per individual per hour, which would be unattainable with manual annotation.”

The researchers found significant genetic variation for all three traits, meaning that part of the differences between the hens can be explained by genetics rather than only by their environment. This is important information, as it means that behavioural traits derived from automated tracking could be used in breeding programs. Moreover, it’s particularly relevant for animal welfare. While management strategies can reduce harmful behaviours, they are often not sufficient on their own. Including behavioural traits in breeding programs could help reduce the genetic predisposition of animals to express such behaviours, complementing management improvements.

Tzayhri and her colleagues are positive about the ArUco markers: “Our study shows that automated tracking using ArUco markers can effectively capture individual behavioural variation with an underlying genetic component, enabling the estimation of breeding values for previously inaccessible traits. Despite technical challenges that led to data loss and modest predictive accuracies, this methodology proved sufficiently robust and powerful to significantly detect genetic variance without major bias. However, the current implementation is hindered by high manual effort, limiting its immediate practicality. Future work must focus on simplifying the entire workflow, from data collection to analysis, to enhance scalability. When refined, this approach provides a scalable framework for integrating detailed behavioural phenotypes into genetic evaluations.”

Read the full study here: From pixels to breeding values: genetic analysis of detection and movement traits in laying hens using automated tracking data of ArUco-marked birds