A computer vision system can be used to help poultry producers manage welfare and production behaviors.
A computer vision system that involves a cameras evaluating the side and top views of a flock can assist poultry producers in monitoring broiler welfare and production related behaviors.
The system uses deep learning technology and algorithms to automatically identify welfare- and production- related behaviors, including lameness, preening, stretching, eating and drinking, allowing producers to better manage birds and compare flocks and management practices.
According to Hao Gan, University of Tennessee Assistant Professor, the research project involved cameras placed on the walls to evaluate lameness and cameras placed on the ceiling to evaluate the remaining behaviors.
Cameras are placed on the walls of the house to achieve a horizontal view of the flock and assign gait scores to broilers, Gan explained at the 2024 Georgia Precision Poultry Farming Conference.
After camera footage is collected, deep learning technology assigns gait scores to individual birds and flocks. The deep learning technology identifies key points on a broiler and generates a skeleton to autonomously detect lameness and assign a gait score with 97.5% accuracy in a research setting, he said.
Gait scores are measured on a scale of 0 – 3, with 0 indicating a bird with no lameness and 3 indicating a medium amount of lameness. Higher scores exist but were not used in the research due to the lab conditions, he added.
The cameras placed on the ceiling of the house achieve an overhead view and involve similar deep learning technology to evaluate how much time birds are spending eating and drinking, as well as the frequency of preening and stretching movements.
Top view cameras must be used to measure production behaviors as the birds grow larger and it becomes more difficult to distinguish them during eating and drinking, he said.
During the research project, the camera systems measured the overall feeding and drinking time with 97.9% and 85.4% accuracy, respectively. The frequency of stretching and preening behaviors was measured with 88.1% accuracy.
More research needs to be done to improve the system’s classification of preening behaviors, he explained.
The project is an extension of previous research involving another camera system to monitor commercial broilers at both the individual and flock level, which was named one of the Phase 1 SMART Broiler winners in 2020.