This work shows the procedure of acquiring and storing the records captured at a fixed sample rate of 10 sps by an IoT collar designed for cows and using video records to identify and tag the respective movements. By tagging, we mean the process of searching, classifying, and manually marking the start and end times of a cow's movement of interest. Thus, the resulting database comprises the temporal signals from the IoT collar's accelerometers and their respective tags, classifying the type of movement the cow performs. This database is a core result in the way for training automatic classifiers for the automatic detection of cow behavior, adding up to a total of 182 hours of registers from 32 cows which were manually classified between walking, resting, sprinting, mounting among others movements of interest.