Non-parametric belief propagation for mobile mapping sensor fusion

Figures & data

Figure 1. Our rigid body node contains internal function nodes (blue) and variable nodes (red). Other nodes connect similarly to the green function node (left).

Figure 2. Illustration of graph generated using our technique.

Note: Rigid body nodes in blue and sensor measurement nodes in orange.

Table 1. Standard deviations used for simulated data.

Parameter	x	y	z
GNSS (m)	1.000	1.000	2.000
Accelerometer (m s^−2)	0.030	0.030	0.030
Gyroscope (deg s^−1)	0.055	0.055	0.055

Table 2. Calibration parameters for sensors.

Parameter	α	β	γ	OX	OY	OZ
Accelerometer	1.0023	0.9904	0.9888	0.1149	0.0889	0.5064
Gyroscope	1	1	1	0.0019	−0.0032	0.0017

Table 3. Measured standard deviation (SD) of sensors

Parameter	SD before calibration			SD after calibration
Axis	x	y	z	x	y	z
Accelerometer	0.0206	0.0200	0.0301	0.0206	0.0198	0.0298
Gyroscope	0.0123	0.0557	0.0054	–	–	–

Figure 3. Performance vs. number of threads.

Figure 4. Scatter plot of calculated positions of the sensor platform x-axis (horizontal) y/z-axis vertical.

Note: X-Y positions are shown as X marker and X-Z positions are shown as + marker.

Figure 5. Plot of calculated positions for real data.

Note: x-axis (horizontal) y/z-axis vertical. X-Y positions are shown as X marker and X-Z positions are shown as + marker.

Figure 6. Plot of calculated positions for real data.

Note: x-axis (horizontal) y/z-axis vertical. X-Y positions are shown as X marker and X-Z positions are shown as + marker.