We know that different brands of lidar produce different data, so how do these differences affect the effectiveness of mapping?

This article is to analyze this problem, will be from different light intensity point cloud effect, different Angle point cloud effect, and

1. Technical index of lidar

Lidar has several indicators:

  • The maximum and minimum distances of the data and the Angle range of the scan (field of view)
  • Angle between 2 data points (angular resolution, points)
  • The frequency at which data is emitted
  • The energy intensity of the data emitted
  • The jump of the point chengdu (accuracy) and so on

1.1 Field of vision

The scope of view of the lidar determines the application scenario of this radar. It is impossible to use the radar with the field of view of 4 meters in a room with the wall distance of 20 meters, and it can not sweep things at all.

1.2 Angular resolution

The Angle resolution of the radar determines the number of data points per frame of the lidar. If it’s a 0.25-degree resolution, 360-degree view of lidar, that frame has a maximum of 1441 points. The more points per frame, the better the representation of environmental details. Of course, the more points is not the better, the more points, the higher the calculation cost.

1.3 the frequency

The frequency of radar is a very important index. The higher the frequency of radar, the smaller the interval between two frames of radar data. Suppose the frequency of radar is 20Hz, that is, 50ms to data. When I do the positioning, the motion of the robot within the 50ms time can only be obtained by estimation. Therefore, the higher the radar frequency is, the smaller the estimated distance will be, and the more accurate the positioning will be.

1.4 Data strength

The laser point of lidar has energy, and different brands of laser point have different energy. When the energy is too small, the data may not be returned at a long distance.

1.5 Data accuracy

This is the most important indicator. If a lidar’s data jumps too much, it’s not going to work. Now the average manufacturer’s radar accuracy is 2%. So at 100 meters, the jump of the point is 2 centimeters. However, there are not many radars that actually feel that accurate.

2 Experimental Analysis

Since we’ve done this before, I’ll just show you the screenshots.

2.1 Point cloud effect under 30000lux light intensity

30000lux is measured with an illuminance meter.

When the sunlight of 30000lux shines on the wall, the lidar will look at the wall illuminated by the sunlight and compare the point cloud effect at this time.

The point cloud effect of pepperl + Fuffer radar is very good. First, it is put out as an experimental comparison term.

We can see the silane radar. It’s not coming back.

The missing part of the data on the right is not due to the light intensity, but due to the insufficient intensity of the laser point. When the Angle is too large, the intensity of the returned data is very small, resulting in no data.

Again, this time the results were even more striking.

The point cloud effect of radium radar on the wall illuminated by the light intensity of 30000lux is still very good.

The circle on the right is also missing data, and it is also speculated that the energy intensity of the laser point is not enough.

The view of the radar is 270 degrees, but only 180 degrees are left due to the robot, so here is its point cloud effect. As you can see, 30,000 lux didn’t bother him.

Beiyang radar’s viewing Angle is 270 degrees, and the light intensity of 30,00lux does not affect him.

2.2 Influence of radar frequency on data distortion

The edges of the map are constructed and rotated to determine whether the radar is distorted by the generated map. When the radar data does not correspond to the map, or the map is distorted, the radar data is distorted.

First of all, we put in the 40Hz Beyang radar, and the figure below shows the data when it is stationary and when it rotates counterclockwise at 1rad/s, and the map built under the rotation.

As you can see, the map is pretty good at this point, and the radar data fits the map pretty well.

The image below shows how it works when rotated counterclockwise, which is also good.

The following figure for pepperl + Fuchs radar 30Hz when the building effect, is also very good.

The following figure shows the data of the Silan radar rotated at 20Hz. Because the energy intensity of the laser point is not high, the data returned by radar is not good enough when the Angle between the wall and radar is too large.

The image below shows the rotation of SICK TIM571 at 15Hz. As you can see, the radar data distortion is very significant. As a result, the map was significantly deformed.

The following figure shows the rotation effect of a 10Hz radium radar, and it can be seen that its distortion amplitude is larger. Probably because the algorithm threw away the distorted radar data, so the map is ok…

The rotation of the radar itself is directional. Most radars rotate counterclockwise, as stipulated in THE ROS, while a few rotate clockwise, which is a little inconvenient to use.

Due to the concern that the rotation direction of the radar itself is the same or opposite to the rotation direction of the robot, which will lead to different effects of lidar data distortion, two kinds of clockwise and counterclockwise rotation are carried out in each group of experiments, but the difference is not obvious after the actual test.

2.3 the conclusion

conclusion

This paper briefly analyzes the different effects of lidar data in different situations.

The article will beOfficial account: Build SLAM from scratchTo synchronize the update, welcome to follow. At the same time, I also hope that you can recommend this official account to the people around you who do laser SLAM, so that we can make progress together.