Estimation of cow's body condition score through statistical shape analysis and regression machines from images acquired using low-cost digital cameras

Body condition score is an indicator of cows’ health status based on visual or tactile inspection. Human assessment of body condition score is the main limiting factor as it is subjective and requires time and well-trained experts. The objective of this study was to explore the potential for using computer vision to assist human experts in this task and for efficient automation of the process of quantitatively estimating the body condition score of cows based on a 5-point scale, using images acquired with commercial low-cost digital cameras. Images were acquired using a camera mounted on a portable device 3 m above the ground, placed in a position which made it possible to capture images of the dorsal area of cows. The body condition score of each cow was estimated on site by 2 technicians and properly associated with the cows’ images. Cow shapes were extracted from the images automatically and aligned in a unique reference frame. Standard principal component analysis was applied to determine the components describing the many ways in which the body shape of different cows tends to deviate from the average shape. The proposed method was tested on a benchmark data set containing 286 images by means of the leave one out cross validation procedure. The error of the proposed method was compared to the performance of other estimation methods based on image evaluation which are reported in the literature. The experimental results confirmed the effectiveness of the proposed technique (error=0.26 body condition score points against human observation) which outperformed other state-of-the-art approaches proposed in the context of dairy cattle research.