Applied Animal Behaviour Science 145 (2013) 44–50 Contents lists available at SciVerse ScienceDirect Applied Animal Behaviour Science j o ur nal homep age : w w w.elsevier.com/locate/applanim In search of the behavioural correlates of optical flow patterns in the automated assessment of broiler chicken welfare Marian Stamp Dawkins a,∗ , Russell Cain a , Kathryn Merelie a , Stephen J. Roberts b a University of Oxford, Department of Zoology, South Parks Road, Oxford OX1 3PS, UK b University of Oxford, Department of Engineering Science, Parks Road, Oxford OX1 3PJ, UK a r t i c l e i n f o Article history: Accepted 3 February 2013 Available online 6 March 2013 Keywords: Broiler chicken welfare Hockburn Pododermatitis Gait Optical flow Behaviour a b s t r a c t Assessment of chicken welfare using camera surveillance of behaviour has great potential as a supplement to good stockmanship and as an aid to improving flock management. Suit- able cameras are now readily available and the automated analysis of the lengthy video sequences from the cameras can be accomplished from statistical descriptors of the ‘opti- cal flow’ patterns produced by flock movements. However, although optical flow measures have been shown to correlate with welfare outputs such as mortality, hockburn, pododer- matitis and gait score, it is not yet clear how the optical flow algorithms achieve this – i.e. what variation in behaviour of individuals the cameras are picking up. The aim of this paper is to clarify the relationship between optical flow, behaviour and welfare by correlating optical flow (mean, variance, skew and kurtosis) taken from 15 min video sequences of commercial broilers at 25 days of age with bird behaviour recorded from the same sequences and with welfare measures from the same flocks (n = 24). There were no significant between-flock correlations (r 2 ) between mean optical flow and % birds walking, % sitting/lying, walking speed or an estimate of the numbers of birds walking continuously for 10 s. or longer (p > 0.02, 2-tailed). Mean % birds walking was, however, significantly and negatively correlated with hock burn (r 2 = −0.44 p < 0.05). Skew and kurtosis of optical flow showed no correlation with % birds walking, % birds sitting/lying or with walking speed (p > 0.02) but were significantly and positively correlated with the number of birds walking continuously for at least 10 s (for skew, r 2 = 0.51, p < 0.01; for kurtosis, r 2 = 0.46, p < 0.05). We conclude that there is no simple connection between optical flow and behaviour or between the behaviour recorded here and mortality, hockburn, pododermatis and gait score. However, reduced numbers of birds walking continuously for 10 s or more appear to be responsible for the increased kurtosis of flow in flocks with higher (less good) gait scores. Optical flow measures correlate more strongly these welfare outcomes than any single behavioural measure. © 2013 Elsevier B.V. All rights reserved. 1. Introduction Continuous, automated monitoring of commercial broiler chicken (Gallus gallus) flocks is potentially a major ∗ Corresponding author: Tel.: +44 01865 271215; fax: +44 01865 310447. E-mail address: marian.dawkins@zoo.ox.ac.uk (M.S. Dawkins). way of improving both welfare and production, not as a substitute for good stockmanship, but as an objective supplement to current methods that are often applied post- mortem (Allain et al., 2009) or are very labour intensive (e.g. gait scoring (Kestin et al., 1992)). With the many thousands of chickens found in modern broiler houses, remote sensing with cameras is a promising route for continuous monitor- ing on commercial farms. But while the required camera equipment is now inexpensive and readily available, there 0168-1591/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.applanim.2013.02.001