A cost-effective imaging system for monitoring poultry behaviour in small-scale kenyan poultry sheds
¹Harper Adams University, Engineering Department, TF10 8NB Newport, United Kingdom
²Harper Adams University, Agriculture and Environment Department, TF10 8NB Newport, United Kingdom
The objective of this paper was to develop a low-cost prototype poultry behaviour imaging and analysis system for monitoring intensively-reared flocks suitable for small-scale Kenyan poultry sheds. An image processing and analysis programme was developed using Python programming language and the OpenCV image processing package. This was tested on overhead images of Ross 308 birds collected over a number of days using a Raspberry Pi V2 camera. A second experiment using toy-chicks was conducted with an angled camera (Wansview W3). Linear transformation (LT) and background subtraction (BS) methods were applied and compared for effectiveness at detecting yellow and brown toy-chicks on woodchip bedding. Perspective transformation (PT) was applied and evaluated for its ability to transform the angled images into two-dimensional views. In the first experiment, where white birds were detected against a dark background, LT object detection successfully detected 99.8% of birds in the sampled images. However, in the second experiment, the LT method was just 56.5% effective at detecting the yellow toy-chicks against the light-coloured background. In contrast, the BS method was more effective, detecting 91.5% of the yellow toy-chicks. The results showed that BS detection success was worse for yellow toy-chicks in the far section, detecting 83% as opposed to 100% of those in the near-section. Edge processing of the image processing algorithm was tested on a Raspberry Pi 3 series B+ computer. This prototype provides a solid foundation for further development and testing of low-cost, automated poultry monitoring systems capable of reporting on thermal comfort inferred from cluster index.