Methodology:
The methodology outlined in this section is an example of the calibration experiment. This methodology must be followed for each unique lidar sensor that is to be simulated as the computed reflectivity value depends on the lidar settings. In this way, the reflectivity conversion is calibrated for each sensor. This is similar to how the value is computed for real lidar sensors, with the calibration happening for each sensor before it leaves the factory.
The calibration experiment consists of two point cloud lidar (PCL) sensors placed at 5 meters distance from two 25 m2 targets (Fig. 1).
Fig. 1: Depiction of the placement of lidars 5 meters away from the targets with diffuse and specular textures.
The sensor configurations are shown in Fig. 2.
Fig. 2: Sensor configuration.
One of the targets was given a perfectly diffuse material, while the other was given a specular texture. These targets were chosen in order to differentiate between them with the reflectivity value. That is: values between 0 and 100 are expected to be diffuse surfaces, while values between 101 and 255 are expected to be specular or retroreflective surfaces. This is in line with the reflectivity value provided by Velodyne sensors.
Next, the power returned by each target was recorded by the sensors (Fig. 3). We use the center beam as its incidence is perpendicular to the target. This returns the highest reflected power from the target which is required for this calibration.
Fig. 3: Power returned from the target with a diffuse texture (left) and the target with a specular texture (right).
Scaling:
We want to find a computation to normalize the power returned by each beam shot by a lidar in order to convert from power to calibrated reflectivity. We want this value to be independent of the laser power and distance to the target. Fig. 4 shows a beam of light illuminating a target located 5 meters away.
Fig. 4: A beam of light illuminating a plane located r meters away.
The power received by the sensor for the calibration experiment,
, is multiplied by the square of the distance to the calibration target,
, in order to remove the effect of distance from the reflectivity computation. This results in a calibrated reflectivity,
:
.
We do this computation for both the diffuse and the specular targets, resulting in the two values
and
. Using these two values we can compute the normalized reflectivity for an arbitrary beam by first computing the measured reflectivity as
.
In order to convert to normalized reflectivity we compute:
,
or:
,
or:
.
This results in a piecewise linear function which maps returned power to normalized reflectivity for any beam.Results:For the calibration example shown in Fig. 1, the values of
obtained from the diffuse target was 0.0019 [W.m2] and for the specular target it was 0.0073 [W.m2]. For a sensor with the configuration shown in Fig. 2, for diffuse materials,
is expected to vary between [0, 0.0019], while for specular materials it is expected to vary between (0.0019, 0.0073]. Any values above this range are expected to be saturating the sensor, and thus are clamped to a normalized reflectivity value of 255. This can occur, for example, for retroreflective targets.
Fig. 5, depicts the normalized reflectivity versus
for the laser considered in this study.
Fig. 5: Normalized reflectivity versus
.