Limits...
Improving the accuracy of estimates of animal path and travel distance using GPS drift ‐ corrected dead reckoning

View Article: PubMed Central - PubMed

ABSTRACT

Route taken and distance travelled are important parameters for studies of animal locomotion. They are often measured using a collar equipped with GPS. Collar weight restrictions limit battery size, which leads to a compromise between collar operating life and GPS fix rate. In studies that rely on linear interpolation between intermittent GPS fixes, path tortuosity will often lead to inaccurate path and distance travelled estimates. Here, we investigate whether GPS‐corrected dead reckoning can improve the accuracy of localization and distance travelled estimates while maximizing collar operating life. Custom‐built tracking collars were deployed on nine freely exercising domestic dogs to collect high fix rate GPS data. Simulations were carried out to measure the extent to which combining accelerometer‐based speed and magnetometer heading estimates (dead reckoning) with low fix rate GPS drift correction could improve the accuracy of path and distance travelled estimates. In our study, median 2‐dimensional root‐mean‐squared (2D‐RMS) position error was between 158 and 463 m (median path length 16.43 km) and distance travelled was underestimated by between 30% and 64% when a GPS position fix was taken every 5 min. Dead reckoning with GPS drift correction (1 GPS fix every 5 min) reduced 2D‐RMS position error to between 15 and 38 m and distance travelled to between an underestimation of 2% and an overestimation of 5%. Achieving this accuracy from GPS alone would require approximately 12 fixes every minute and result in a battery life of approximately 11 days; dead reckoning reduces the number of fixes required, enabling a collar life of approximately 10 months. Our results are generally applicable to GPS‐based tracking studies of quadrupedal animals and could be applied to studies of energetics, behavioral ecology, and locomotion. This low‐cost approach overcomes the limitation of low fix rate GPS and enables the long‐term deployment of lightweight GPS collars.

No MeSH data available.


Related in: MedlinePlus

The relationship between GPS speed (mean over a 2‐s window) and the accelerometer speed proxy (VeDBA) for domestic dogs. (A) An example from dog 2. VeDBA was calculated over a 2‐s window. The linear fit to these data (training data) and 95% confidence intervals (CI) is represented by the solid red and dash dotted lines, respectively (r2 = 0.74). We use the distribution of the GPS speeds and VeDBA values (histograms with kernel density estimates marked by black lines) to determine areas on the scatter plot which represent walking, trotting, and running gaits. The local minimum of the kernel density estimate is taken as the transition point between walking and trotting. (B) Scatter plot showing the relationship between GPS speed and VeDBA for all dogs and the linear fit for each dog (D1 to D9).
© Copyright Policy - creativeCommonsBy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC5016644&req=5

ece32359-fig-0001: The relationship between GPS speed (mean over a 2‐s window) and the accelerometer speed proxy (VeDBA) for domestic dogs. (A) An example from dog 2. VeDBA was calculated over a 2‐s window. The linear fit to these data (training data) and 95% confidence intervals (CI) is represented by the solid red and dash dotted lines, respectively (r2 = 0.74). We use the distribution of the GPS speeds and VeDBA values (histograms with kernel density estimates marked by black lines) to determine areas on the scatter plot which represent walking, trotting, and running gaits. The local minimum of the kernel density estimate is taken as the transition point between walking and trotting. (B) Scatter plot showing the relationship between GPS speed and VeDBA for all dogs and the linear fit for each dog (D1 to D9).

Mentions: The scatter plot in Figure 1A shows a representative example of the relationship between GPS speed (mean over a 2‐s window) and VeDBA (domestic dog 2) and a linear fit to the data (r2 = 0.74). The distribution of the GPS speed and VeDBA values (histograms and kernel density estimates to the right and top of the scatter plot) is used as a guide to show which areas of the scatter plot represent walking, trotting, and running gaits (Maes et al. 2008). The local minimum of the kernel density estimate is used to represent the gait transition between walking and running (Fig. 1A). The bimodal distribution of the GPS speed measurements (Fig. 1A) suggests that domestic dog 2 is using preferred speeds (Hoyt and Taylor 1981) within its walking and trotting gaits. Similar results were found in the other dogs.


Improving the accuracy of estimates of animal path and travel distance using GPS drift ‐ corrected dead reckoning
The relationship between GPS speed (mean over a 2‐s window) and the accelerometer speed proxy (VeDBA) for domestic dogs. (A) An example from dog 2. VeDBA was calculated over a 2‐s window. The linear fit to these data (training data) and 95% confidence intervals (CI) is represented by the solid red and dash dotted lines, respectively (r2 = 0.74). We use the distribution of the GPS speeds and VeDBA values (histograms with kernel density estimates marked by black lines) to determine areas on the scatter plot which represent walking, trotting, and running gaits. The local minimum of the kernel density estimate is taken as the transition point between walking and trotting. (B) Scatter plot showing the relationship between GPS speed and VeDBA for all dogs and the linear fit for each dog (D1 to D9).
© Copyright Policy - creativeCommonsBy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC5016644&req=5

ece32359-fig-0001: The relationship between GPS speed (mean over a 2‐s window) and the accelerometer speed proxy (VeDBA) for domestic dogs. (A) An example from dog 2. VeDBA was calculated over a 2‐s window. The linear fit to these data (training data) and 95% confidence intervals (CI) is represented by the solid red and dash dotted lines, respectively (r2 = 0.74). We use the distribution of the GPS speeds and VeDBA values (histograms with kernel density estimates marked by black lines) to determine areas on the scatter plot which represent walking, trotting, and running gaits. The local minimum of the kernel density estimate is taken as the transition point between walking and trotting. (B) Scatter plot showing the relationship between GPS speed and VeDBA for all dogs and the linear fit for each dog (D1 to D9).
Mentions: The scatter plot in Figure 1A shows a representative example of the relationship between GPS speed (mean over a 2‐s window) and VeDBA (domestic dog 2) and a linear fit to the data (r2 = 0.74). The distribution of the GPS speed and VeDBA values (histograms and kernel density estimates to the right and top of the scatter plot) is used as a guide to show which areas of the scatter plot represent walking, trotting, and running gaits (Maes et al. 2008). The local minimum of the kernel density estimate is used to represent the gait transition between walking and running (Fig. 1A). The bimodal distribution of the GPS speed measurements (Fig. 1A) suggests that domestic dog 2 is using preferred speeds (Hoyt and Taylor 1981) within its walking and trotting gaits. Similar results were found in the other dogs.

View Article: PubMed Central - PubMed

ABSTRACT

Route taken and distance travelled are important parameters for studies of animal locomotion. They are often measured using a collar equipped with GPS. Collar weight restrictions limit battery size, which leads to a compromise between collar operating life and GPS fix rate. In studies that rely on linear interpolation between intermittent GPS fixes, path tortuosity will often lead to inaccurate path and distance travelled estimates. Here, we investigate whether GPS‐corrected dead reckoning can improve the accuracy of localization and distance travelled estimates while maximizing collar operating life. Custom‐built tracking collars were deployed on nine freely exercising domestic dogs to collect high fix rate GPS data. Simulations were carried out to measure the extent to which combining accelerometer‐based speed and magnetometer heading estimates (dead reckoning) with low fix rate GPS drift correction could improve the accuracy of path and distance travelled estimates. In our study, median 2‐dimensional root‐mean‐squared (2D‐RMS) position error was between 158 and 463 m (median path length 16.43 km) and distance travelled was underestimated by between 30% and 64% when a GPS position fix was taken every 5 min. Dead reckoning with GPS drift correction (1 GPS fix every 5 min) reduced 2D‐RMS position error to between 15 and 38 m and distance travelled to between an underestimation of 2% and an overestimation of 5%. Achieving this accuracy from GPS alone would require approximately 12 fixes every minute and result in a battery life of approximately 11 days; dead reckoning reduces the number of fixes required, enabling a collar life of approximately 10 months. Our results are generally applicable to GPS‐based tracking studies of quadrupedal animals and could be applied to studies of energetics, behavioral ecology, and locomotion. This low‐cost approach overcomes the limitation of low fix rate GPS and enables the long‐term deployment of lightweight GPS collars.

No MeSH data available.


Related in: MedlinePlus