Long-term monitoring of animal activity can yield key information for both researchers in ethology and engineers in charge of developing precision livestock farming tools. First, a barn is segmented into delimited areas (e.g. cubicles) with which an activity can be associated (e.g. resting), then a real-time location system (RTLS) can be used to automatically convert cow position into behaviour. Working within the EU-PLF project, we tested a system already able to determine basic activities (resting, moving, eating…) and logged a “big data” set of billions of data points (123 days 190 cows 1 location-per-second readings). We then focused on integrating image analysis techniques to help visualise and analyse the dataset, first to validate the data and then to enrich the information extracted. The algorithm developed using freely available tools quickly confirmed the ability of the system to determine cows' main activities (except drinking behaviour), even with 11% of positions missing. The good localisation precision (16 cm) made it possible to enrich the time-budget with new activities such as using brushes and licking mineral blocks. For both activities, using visual observations as gold standard, activity profiles with excellent sensitivity (nearly 80%) were extracted. This validation procedure is both necessary and generalisable to other situations. The improvement of biological information contained in such data holds promise for people designing alarm devices and health and welfare indicators for farmers and/or vets.