The ability to measure individual feed intake has been a desire from the dairy industry for long, in order to breed for feed efficiency, but also to support farm management decisions. In addition, changes in feed intake and feeding behaviour are indicators of disease (e.g. ketosis, lameness) in dairy cows. Roughage intake control bins are often used to measure the feed intake of individual cows in research. However, this method is not applicable on commercial farms due to its high investment cost. Computer vision based solutions may serve as an alternative that can be applied under practical farm circumstances at relatively low investment cost.
We will develop and validate a computer vision system for feed weight measurements based on recordings from an RGB-depth camera in a dairy barn. These cameras are able to record the three-dimensional properties of objects in addition to capturing a normal (RGB) video as we know it. An RGB-depth camera will be installed above the feed pile in a dairy barn, and feed weight will be measured using an automatic scale in parallel. Coupling the feed volume and other characteristics from the top-view recordings with machine learning and deep learning algorithms, we will estimate the weight of known amounts of feed.
Once the method for feed weight estimation is validated in our current experiment, it will be tested with cows to estimate daily individual feed intake. Using the same recordings, this method also allows for more detailed behavioural observations (e.g. feeding pattern, aggression), so the current experiment is just the first step of a long and exciting journey.