The performance indicators of lidar for autonomous driving sensors

In the last issue, we briefly learned about the main components of vehicle-mounted sensors and a brief introduction to vehicle-mounted laser radar. Many friends said that they can distinguish the major categories, but they still don’t know some specific functional parameters. So in this issue, we will introduce the performance indicators of laser radar.

Let me give you a small test first.

For example, the picture below is a parameter table of a 128-line laser radar product. Can you understand it?

Image source: RoboSense official website

It doesn’t matter if you don’t understand. I believe that after reading our article today,

You can understand it. Let's talk about them one by one.

1. Number of lines

Car-mounted LiDAR generally uses multi-line radar, which means it has multiple transmitters and receivers in the vertical direction. For LiDAR, more and denser lines mean clearer and more complete collection of 3D contours of various objects. Currently, most autonomous vehicles use 16, 32, 64, 128 lines, etc.

In scenarios such as surveying and autonomous driving, the higher the number of laser radar lines, the better, but it also has a disadvantage.

02. Wavelength

Currently, there are two most commonly used wavelengths for three-dimensional imaging lidar on the market, namely 905nm and 1550nm.

Comparing the two, the indium gallium arsenide (InGaAs) near-infrared detector used in the 1550nm wavelength laser radar is more expensive than the silicon-based photodetector used in the 905nm wavelength, so there will be a difference in the cost of the two. In fact, you get what you pay for. The laser power of the 1550nm laser radar is 40 times that of the traditional 905nm silicon photoelectric system; it also effectively solves the problem of insufficient penetration of 905 nanometers in rain and fog, so the effective detection distance can be maintained at more than 200 meters in complex climate and weather conditions.

At the same time, lasers with a wavelength of 1550nm are also safer and harmless to the human eye. The reason is that in order for the lidar to "see farther", it is necessary to emit stronger light pulses. However, the longer the wavelength, the lower the energy and the farther it is from visible light. 905-nanometer lasers are invisible to humans. But if it is focused on a real person's eyeball and has enough energy, it will damage the human retina. In contrast, the 1550-nanometer laser is farther away from the visible light band in the infrared part of the spectrum, so it cannot penetrate the human eyeball.

03. Laser safety level

The above mentioned the harm of laser to human eyes. In fact, the International Electrotechnical Commission (IEC) has formulated safety standards. It divides laser products into 7 safety levels based on the wavelength, maximum output laser power or energy of laser products. Automotive laser radar generally requires the human eye safety level to be Class 1.

04. Ranging capability and blind spots

Let's talk about these two together. First, the ranging capability (range), that is, the ability to effectively detect the nearest and farthest distances. In addition, the higher the reflectivity of the target, the longer the measurement distance, and the lower the reflectivity of the target, the closer the measurement distance.

For example, "200m@10% reflectivity" in the parameter table means that the laser radar can emit a light beam to a distance of 200 meters and can clearly see objects with a minimum light reflectivity of 10%, such as a piece of white paper.

Image source: RoboSense official website

Blind spot, as the name implies, is the distance where the laser radar cannot receive the laser echo signal. In a word, it is the blind spot outside the visible range. It feels a bit like the fog of war in the game scene.

Image source: Age of Empires

Blind spots can also occur due to close distances. For example, in the figure below, when an object is close to the lidar hood, lidar products with good performance can still detect it effectively and achieve "zero blind spots", while some lidars have blind spots at 0.2 meters and cannot form images.

Comparison of laser radar blind spots (Image source: Hesai Technology official website)

By the way, for NavInfo, although the ranging capability (range) of LiDAR is different, NavInfo's autonomous driving solution can allocate the best solution according to the characteristics and indicators of different sensors to ensure the accuracy and robustness of perception. For example, the LiDAR can be installed in certain positions and fixed angles of the vehicle body, and then the problem can be effectively solved through NavInfo's own sensor fusion algorithm. This solution has passed the Beijing Autonomous Driving Road Test T3 license and the Hefei Intelligent Connected Vehicle Open Road Test license.

NavInfo self-driving car

05. Accuracy

Accuracy is the minimum change in distance that the radar can perceive. Generally speaking, the farther the target is from the lidar, the lower the accuracy of the distance measurement. For example, the human eye can clearly see an object the size of a wine bottle cap at a distance of 10 meters, but it is invisible after 200 meters. As shown in the figure below, the number of lidar point clouds decreases as the distance of the car object decreases, and the accuracy of the car body boundary contour decreases accordingly.

Image source: RoboSense official website

06. Field of View

Field of View (FOV) is the effective scanning angle within which the target object can be detected. It is divided into horizontal field of view and vertical field of view. Taking the human eye as an example, it is the up and down and left and right range that you can see when you look straight ahead.

The horizontal field of view is the range detected on the horizontal plane. Taking a 64-line laser radar as an example, the range of a simple 64-line laser beam is far from enough, but if the range of the beam is increased, the cost will be greatly increased. Therefore, a mechanical rotary laser radar with a motor is born, which can form a 360° horizontal field of view with the rotation action, while a solid-state laser radar can only see certain degrees directly in front (for example, 60° to 120°);

VelodyneHDL-64E horizontal scanning diagram

The vertical field of view angle and vertical azimuth angle refer to the detection angle of the laser radar in the vertical direction, which is generally within 40°. Moreover, the light beam is not evenly distributed vertically, but dense in the middle and sparse on both sides. In addition, the laser beam will try to deflect downward at a certain angle, which is also to scan more obstacles and to a certain extent make up for the near blind spot mentioned above.

VelodyneHDL-64E vertical scanning diagram

07. Angular resolution

Angular resolution refers to the angle between two adjacent ranging points. As shown in the figure below, it is also divided into horizontal angular resolution and vertical angular resolution. Like the FOV above, horizontal angular resolution and vertical angular resolution respectively represent the angle between points on the horizontal plane and the angle between points on the vertical plane. I will not go into details here.

Angular resolution diagram

In addition, the angular resolution determines the number of sample points that can be returned in one scan and the minimum obstacle size that the laser radar can detect. For example, if the angle between two adjacent ranging points of laser beams is 0.4°, then when the detection distance is 200m, the distance between the two laser beams is 200m*tan0.4°≈1.4m. That is to say, after 200m, only obstacles higher than 1.4m can be detected.

Image source: RoboSense official website

08. Frame rate

Frame rate and scanning frequency indicate how many times the radar scans in one second. The unit of measurement is HZ (Hertz is the unit of frequency of the cycle of electric, magnetic, sound waves and mechanical vibration. That is, the number of cycles per second "cycles/second"). The higher the scanning frequency, the faster the device perceives the external environment, and the higher the real-time performance of the system.

09. Speed

The international standard unit of rotation speed is rps (revolutions per second) or rpm (revolutions per minute). 1200rpm means that the motor engine speed of the current laser radar is 1200 revolutions per minute. Like the frame rate, the higher the rotation speed, the faster the device perceives the external environment.

10. Points

The number of output points (divided into single echo mode and double echo mode) refers to how many points the laser radar can scan per unit time.

Let's take a 128-line mechanical rotating laser radar as an example to calculate the number of output points:

A known:

Mechanical rotating laser radar has a 360° horizontal field of view

10Hz frame rate (scans 10 times per second), which means it rotates 360 degrees every 100ms (10Hz)

Horizontal angular resolution 0.2°

Therefore, when the horizontal angular resolution is 0.2°, one rotation will generate 360/0.2=1800 points.

Finally, the output point number is 128 lines * 1800 points * 10HZ = 2304000 points/second (pts/s)

11. Other parts

For car manufacturers, the power supply voltage, product power, product operating temperature, IP dust and feng shui protection levels of all electrical components in the car are very important purchasing indicators, which are not what we need to pay attention to here.

Conclusion:

I believe that everyone will be familiar with the parameter table at the beginning. In fact, LiDAR is not that mysterious. To sum up the performance of LiDAR in one sentence: through the good ranging ability, penetration and rich laser point cloud imaging ability of laser, it can help autonomous driving vehicles see farther and more clearly, and obtain three-dimensional information around the vehicle at any time.

With the rapid commercialization of LiDAR products at home and abroad recently, combined with the progress of Huawei, Hesai and other companies in the field of LiDAR mentioned in the previous article, and the launch time of LiDAR products equipped with autonomous driving vehicles by BAIC, Xiaopeng, Weilai and other car companies, what is the situation of LiDAR abroad? How are the relevant companies in China progressing? In the next issue, the editor will sort out relevant information for you, so stay tuned!