Optical flow, behaviour and broiler chicken welfare in the UK and Switzerland

Highlights

• The welfare of broiler chickens affects the pattern of their flock movements.

• Cameras can ‘read’ these movements to predict future mortality and hockburn.

• Differences between flocks are apparent in chicks of only a few days old.

• Flocks carry their history with them throughout life.

• Results are repeatable in the different conditions of the UK and Switzerland.

Abstract

Although systems for automated assessment of broiler chicken welfare have now been developed on a small scale, none are currently in widespread commercial use. We addressed this gap between research and uptake by field testing a camera system that uses the optical flow patterns made by the movements of flocks to monitor bird welfare. We tested the hypothesis that the movement patterns made by flocks of broiler chickens are correlated with two key welfare outcomes – mortality and hockburn. Life-long CCTV monitoring was carried out on 74 commercial broiler flocks (UK = 31; Switzerland = 43) and the resulting data analysed to give daily values of 4 optical flow descriptors: mean, variance, skew and kurtosis. Flock mortality and hockburn data were obtained from information routinely collected on the farms and at abattoirs. Bayesian multivariate regression models were used to analyse data with all 4 optical flow descriptors as input variables, first day-by-day and then cumulatively using information from each day and all previous days. For both the UK and Swiss flocks, the cumulative regression showed that optical flow was significantly (p < 0.01) correlated with % total mortality (by day 1 for the UK flocks and by day 4 for the Swiss flocks). Optical flow was also significantly (p < 0.01) correlated with % birds with end-of-life hockburn (by day 3 for UK flocks and day 2 for Swiss flocks). This ability to distinguish between flocks that will have different final welfare outcomes when the birds are only a few days old is a particularly useful property of this system as it potentially provides farmers with an early warning of problem flocks that are still young enough for interventions to be possible. In conclusion, the optical flow patterns made by the movements of broiler chicken flocks were reliably correlated with two key welfare outcomes - mortality and hockburn - obtained by cumulating information over days as the birds grow. Differences between flocks became apparent within the first few days of life and were similar in the UK and Switzerland, despite differences in environment and management practices, suggesting that this approach to automated welfare assessment has the potential for widespread commercial use.

Introduction

‘Smart’ or ‘precision’ farming has revolutionized the cultivation of agricultural crops but its application to livestock farming has raised ethical concerns because of its possible adverse effects on animal welfare (Wathes et al., 2008; Banhazi et al., 2012; Fournel et al., 2017; Winckler, 2019). Some people see the efficiency gains offered by the new technology as a direct threat to the animals themselves, allowing producers to get “more for less” in the interests of efficiency and to further encourage the use of intensive systems which they see as damaging to welfare (Stevenson, 2017). Others see major welfare advantages through life-long health monitoring, the ability to deliver care to individual animals and to provide optimal environmental conditions (Wathes et al., 2008; Banhazi et al., 2012; Berckmans, 2017; Buscher, 2019; Veissier et al., 2019). They also point to other potential benefits of the technology such as reducing waste and cutting the need for medication through improved animal health and lowered environmental impact (Clark and Tilman, 2017; Lovarelli et al., 2020; Perakis et al., 2020).

The extent to which smart technology has been adopted by farmers varies greatly with the animals concerned. With dairy cows, for example, changes in the health status and behaviour of each individual cow can now be monitored through a variety of different devices attached to a cow’s body or in the environment (Beer et al., 2016; Alsaaod et al., 2019). As a result, many dairy farmers have found that investment in such technology benefits their entire business (Lovarelli et al., 2020). However, with animals kept in large groups such as laying hens and broiler chickens, there are technological issues arising from the need to monitor many thousands of animals simultaneously and although a number of systems have been developed (Dawkins et al., 2009, 2012; Aydin et al., 2010; Silvera et al., 2017; Fernandez et al., 2018; Li et al., 2020), automated assessment of poultry health and welfare is not currently in widespread commercial use (Rowe et al., 2019; Rios et al., 2020). One major reason for this may be that no system has yet been sufficiently validated in a commercial setting for producers to be confident that it is worth adopting (Wathes et al., 2008; Wurtz et al., 2019; Dawkins and Rowe, 2020). More attention therefore needs to be given to validation and to farmer adoption strategies (Rios et al., 2020).

We here report a study on the welfare of broiler chickens that uses cameras to analyse the optical flow patterns made by flocks of chickens as they move around a house (Dawkins et al., 2012; Roberts et al., 2012). Although some automated systems of welfare assessment focus on the behaviour of individual birds (e.g. Naas et al., 2018), the system tested here is one of an increasing number that focus on behaviour and welfare at group level (Dawkins et al., 2009; Aydin et al., 2010; Ben Sassi et al., 2016; Van Hertem et al., 2018). We analysed the movements of whole flocks by detecting the rate of change of image brightness (‘optical flow’) in different parts of the whole camera image both through time and space (Beauchemin and Barron, 1995; Fleet and Weiss, 2005). These changes were then combined to give an estimate of local velocity vectors.

This approach has already been successfully used on commercial farms in the UK and has demonstrated significant correlations between optical flow parameters and key welfare outcomes for poultry such as hockburn, pododermatitis (Dawkins et al., 2009, 2012; 2017) and infection with Campylobacter (Colles et al., 2016). We here tested the hypothesis that the optical flow patterns of flocks less than two weeks old are correlated with two key welfare outcomes – mortality and % hockburn – measured several weeks later and can therefore provide an early warning to farmers of potential health and welfare problems. By obtaining results from commercial farms in both the UK and Switzerland, we aimed to demonstrate that this approach was practical enough to be deployed on commercial farms even where there are differences in farm conditions and in times and methods of data collection.