Validation of an indoor real-time location system for tracking sheep
Data files
Abstract
Precision livestock technologies such as remote sensors are increasingly used to monitor the health, behavior, and welfare of livestock. We aimed to evaluate the performance of a commercially available ultra-wideband real-time location system (UWB RTLS) for tracking the 2D spatial locations and distances traveled by meat-breed ewes and lambs in an indoor barn. First, we assessed static performance by attaching the sensors to stationary posts and arranging them in a 1 x 1 m grid throughout the barn (29.0 x 11.8 m) for a total of 285 locations. At each post location, the sensors were placed at approximate ewe (0.9 m) and lamb (0.3 m) wither height. The precise 2D locations of each post were recorded using a laser tape measurer and used as the ground truth for comparison to the RTLS’ recorded x and y coordinates. Secondly, we conducted a dynamic validation test to evaluate the positional error and percent error of distances traveled while the sensors were worn by six free-roaming ewes and their singleton lambs. The ground truth locations of each sheep were recorded from video frames every second over 15 minutes and compared to the RTLS data. Overall static and dynamic error was 0.39 ± 0.20 m (mean ± SD) and 0.53 ± 0.31 m, respectively. Static error was lower in sensors positioned at lamb height than at ewe height, but the opposite pattern was true for dynamic error. Error was higher in pens further from the master anchor. Ground truth and RTLS distances traveled were positively correlated but the RTLS overestimated distances by 54% on average. In conclusion, the UWB RTLS can acquire precise location estimates that are suitable for a range of scientific and practical applications, but distance estimates should be adjusted to account for overestimation.
https://doi.org/10.5061/dryad.d7wm37q9b
Description of the data and file structure
These materials provide the data and R scripts used in this experiment and are referenced in the corresponding paper by the same name.
In this experiment, we aimed to test the static and dynamic performance of an ultra-wideband (UWB) real-time location system (RTLS). We assessed static performance by attaching the sensors to stationary posts and arranging them in a 1 x 1 m grid throughout the barn (28.96 x 11.81 m). At each post location, the sensors were placed at approximate ewe (0.9 m) and lamb (0.3 m) wither height to determine whether height affected positional error. Location data were recorded for approximately one minute. The precise 2D locations of each post were recorded using a laser tape measurer and used as the ground truth for comparison to the system’s recorded coordinates. We assessed dynamic performance to evaluate the RTLS positional error and percent error of distances traveled while the sensors were worn by six free-roaming ewes and their singleton lambs. The ground truth locations of each sheep were recorded from video frames every second over 15 minutes and compared to the RTLS data.
Data files on Dryad
Supplementary File S1 (.xlsx): Excel document containing the ground truth and RTLS x,y coordinates for each post position by pen, row, and height.
| Variable | Definition |
|---|---|
| TagId | Unique identifier for each sensor (10 total) |
| Pen | Pen that the data were recorded in (1-3) |
| Row | Row where the post was position (1-11) |
| Height | Height of the sensor on the post (Ewe or lamb wither height) |
| StartTime | Time that the positional data were recorded |
| MeasuredX | Manually measured ground truth x coordinate for each post. Measured as distance in m from the origin point. |
| MeasuredY | Manually measured ground truth y coordinate for each post. Measured as distance in m from the origin point. |
| meanX | RTLS measured x coordinate for each post. Measured as distance in m from the origin point. |
| meanY | RTLS measured y coordinate for each post. Measured as distance in m from the origin point. |
Supplementary File S4 (.xlsx): Excel document containing the ground truth and RTLS x,y coordinates for each sheep for every second over 14.5 minutes by pen.
| Variable | Definition |
|---|---|
| ID | Identification of focal subject |
| PhotoTime | Time that the positional data were recorded |
| Pen | Pen that the data were recorded in (1-3) |
| Breed | Breed of sheep (Hamp = Hampshire; Poly = Polypay) |
| MeasuredX | Ground truth x coordinate for each sheep obtained from video frames. Measured as distance in m from the origin point. Blank cells indicate unrecorded data due to not visible front hooves of the individual. |
| MeasuredY | Ground truth y coordinate for each sheep obtained from video frames. Measured as distance in m from the origin point. Blank cells indicate unrecorded data due to not visible front hooves of the individual. |
| RecordedbySensor | Binary variable indicating whether the meanX and meanY values were recorded by the sensor (1) or filled in post-processing because the sensor was asleep (0). |
| meanX | RTLS measured x coordinate for each sheep. Measured as distance in m from the origin point. |
| meanY | RTLS measured y coordinate for each sheep Measured as distance in m from the origin point. |
| FrontHoof | Binary variable indicating whether the front hooves of the focal animal were visible (1) or not (0) when coding video frames |
Supplementary Files and Software on Zenodo
Supplementary File S2 (nb.html): This R script uses the data file S1 to determine the error of the static performance and generates Figure 2 in the corresponding paper. The nb.html file, which includes annotated code and output, can be opened in any web browser or in RStudio (e.g., using File -> Open File).
Supplementary File S3 (.png): Photo of a lamb wearing the custom-built harness to hold the RTLS sensor in place on the shoulder blades. The black pouch contained the sensor.
Supplementary File S5 (nb.html): This script uses the data file S4 to calculate and analyze the positional error, distance traveled, and percent error and creates Figures 3 and 4 in the corresponding paper. The nb.html file, which includes annotated code and output, can be opened in any web browser or in RStudio (e.g., using File -> Open File).
Supplementary File S6 (.docx): Table displaying the dynamic performance of the RTLS compared to ground truth measurements extracted from video. The mean, standard deviation, minimum, and maximum of the error as calculated using RMSE are shown for each individual.
Please see the description in the associated research publication.