Machine Vision Detection of Range Use by Free-Range Meat Chicken Flocks

Quantification of range use in commercial meat chicken flocks is desirable so that farmers can objectively assess meaningful range access, including number of chickens using the range and how far chickens roam from the shed. Currently there is no cost-effective technology for quantifying range use in commercial conditions, and farm staff perform assessments based on visual inspection. A proof-of-concept machine vision system was developed for automated range use assessment, using simple and low-cost camera hardware with visibility of the length of the range area. A colour camera with solar panel and power bank was used to monitor a free-range site in South Australia in March 2022. The camera was placed at one end of the range adjoining a meat chicken shed, midway across the 12 m width of the range, and oriented towards the far end of the range. The camera was positioned at a height of 1.8 m and programmed to capture one image per minute during daylight hours. Image resolution was 1600 × 1200 pixels. Images were captured for 14 days on which pop holes were open, spanning flock age of 21 to 42 days.

Captured images were a one-point perspective scene, with the edge of the shed and the boundary fence of the range receding to a vanishing point at the middle top of the image. Plant ground coverage was 85%, comprising green kikuyu grass, dry annual grasses, and isolated broadleaf weeds. The shed was oriented east-west such that shadows cast by the shed onto the range were relatively short. However, there were silos behind the camera on the eastern end of the range which cast a long shadow onto the range in the morning. Hence, image analysis was required to detect chicken pixels for sunlit, shaded and overcast conditions. After chicken pixels were detected, number of chickens and their distance from the shed were automatically calculated taking image perspective into account.

Image analysis counts of chickens achieved R2 of 0.85 to 0.94 when compared with manual counts of 120 images across four days with variable natural lighting. Reduced image analysis accuracy occurred within 1 m of the shed where chickens were seated closest together. Image analysis and visual inspection both indicated that range use was highest at approximately 0900 - 0930 h for the observed flock, with range use declining after about 1030 h for sunny days. Based on image analysis, it was common for chickens to range up to 4 m from the shed. Chickens appeared to prefer ranging in the shadow cast by the silos, as chickens that were the greatest distance from the shed (i.e. more than 4 m) were always seated or standing in the shadow cast by the silos, where that same shadow also reached within 3 m of a pop hole.

It was concluded that a single low-cost colour camera with a viewpoint along the range was suitable for image analysis algorithms to quantify range use. Automated counts were accurate compared to manual counts from images. The algorithms could distinguish chickens under the different lighting conditions of overcast, sunlit and shaded, indicating potential for further research of meteorological and environmental conditions and ranging behaviour. Further work should evaluate and refine algorithms for segmenting chickens from the background for various ground covers, sunlight directions and flock ranging routines, through trials for a greater number of farms, shed orientations and times of year.