LambCam
For most sheep farmers, lambing season is a very busy time of year. Besides feeding the sheep and providing bedding, the farming team must monitor the stage of labour of the ewes, pen them, assist with the delivery when needed, and perform care of the newborn lambs. All of this can be very time-consuming over the several weeks that lambing takes place.
In recent years, the use of CCTV by farmers to monitor their flock remotely has made things easier. Besides cameras, other data-gathering technologies have been used to gather a wider range of data to indicate the state of labour of the ewes. The additional information contributes to the welfare, wellbeing and health of the animals, while reducing the risk of exhaustion in the farming team. The LambCam project is a multi-year research and development project underway to use streams of multi-modal data gathered in the farming environment in order to improve outcomes in ovine obstetrics. In doing so, we apply and develop techniques from computer vision, machine learning and other fields.
The work is a result of the collaboration of two sets of experts:
- The farming team at Great Ashley Farm in west Wiltshire. Their flock of eighty breeding ewes are reared traditionally to high standards and are the focus of the studies in this project;
- A group of research technologists at The Open University's Knowledge Media Institute who, along with independent engineers and researchers, are building and developing hardware and software for the project.
Objectives and research questions
- To explore the application of computer vision models for taking a visual scene and converting it to structured data about the flock. For example, how can flocking patterns be characterised and shown to correlate with events in first and second stage labour in sheep?
- To explore the application of machine learning techniques for predicting events relating to labour. For example, how can early-stage labour events be reliably predicted from camera footage of the flock to minimise disturbance to the ewe and otherwise improve outcomes?
- To build and refine low-cost, open source equipment for use in the farm environment to easily and effectively capture the raw data for analysis. For example, IP cameras, microphone arrays, or EID (ear tag) readers.
Progress so far
- We're working on low-cost open-source EID readers to read the tags attached to the ewe's ears. Ordinarily this enables the farmer to identify the animal but for our purposes it also enables data analysts and computer models to isolate profiles of an individual ewe and to predict events and outcomes on a per-animal basis.
- We've designed and built a set of low-cost open source IP cameras and networked them so they monitor the main lambing area and the feed face. We've collected many weeks of video consisting of tens of millions of frames. From this we will characterise the flocking dynamics in order to construct a higher level model of the state of the flock, as well as annote the data with major labour events in each ewe.
To learn more about our progress, take a look at our GitHub organisation or join our mailing list for project updates in the future. We'll also do our best to post updates here too.