Localisation and identification performances of a real-time location system based on ultra wide band technology for monitoring and tracking dairy cow behaviour in a semi-open free-stall barn

•A UWB RTLS was installed in a free-stall barn to acquire locations of dairy cows.•Rectified top-view images of barn allowed visual recognition of cows’ true positions.•Software were specially designed to assess tag positions provided by the RTLS.•RTLS localisation and identification performances were evaluated and balanced.•This RTLS is suitable for cow behavioural indices computation and animal tracking. The objective of this study was to evaluate the localisation and identification performances of a Real-Time Location System (RTLS) based on Ultra Wide Band (UWB) technology within a semi-open free-stall barn since the conditions of the breeding environment were different from that of the ‘typical open environment’ used by the RTLS producer to test the system and the building characteristics were dissimilar to those of the indoor environments considered in other tests. Each dairy cow was equipped with an active tag applied to one ear and a reference tag was fixed to a pillar of the barn. A video-recording system was installed in the barn to perform the assessment of the RTLS. Top-view camera images of the area of the barn were rectified and synchronised with the RTLS. An operator validated each position of the cow computed by the RTLS by performing cow visual recognition on the camera images. To perform this validation a software specifically designed for the purpose was utilised. It is an automatic and interactive tool which includes selection and control tabs for data management, visualisation and labelling of the images with the aim of computing tag true positions. RTLS localisation and identification performances were assessed by applying an outlier data cleaning technique to tag localisation errors and using precision and sensitivity indices. Trade-off between these performances was found through the computation of three performance metrics. The combination between the outlier data cleaning technique and the trade-off analysis of RTLS performances yielded the localisation mean error that was computed by averaging the localisation errors of each tag. It was equal to about 0.11m with an identification accuracy of nearly 100% for the reference tag, whereas for the tags applied to the cows the average localisation mean error, computed by averaging the localisation mean errors of the tags, was about 0.515m with an identification accuracy of 98%. At the 90th percentile the average localisation mean error was about 0.967m for the cows’ tags, whereas it was