LIDAR (a combination of light and radar) is an indispensable component of many autonomous vehicles. It measures distance through light and becomes the eyes of autonomous vehicles.
The working principle of LiDAR
Lidar measures the flight time of light pulses, which can then determine the distance between the sensor and the object. Imagine starting the stopwatch when a light pulse is emitted, and then stopping the timer when the light pulse (reflected from the first object encountered) returns; by measuring the "time of flight" of the laser and knowing the speed of the pulse, you can calculate the distance. Light travels at a speed of 300,000 kilometers per second, so very high-precision devices are required to generate distance information.
Use a laser as a "stick" to measure the distance
To generate a complete point cloud, the sensor must be able to sample the entire environment very quickly. LiDAR uses a very high sampling rate on a single transmitter/receiver, and each transmitter emits tens or hundreds of thousands of laser pulses per second. This means that as many as 100,000 laser pulses complete the round trip from the transmitter on the laser unit to the object to be measured within 1 second, and return to the receiver located near the transmitter on the lidar. Larger systems have up to 64 such transmitters/receivers (it is called a line). Multi-line enables the system to generate more than one million data points per second.
However, 64 fixed lines are not enough to map the entire environment-it just gives a very clear resolution in a very concentrated area. Due to the precision required in optics, it is very expensive to make more wires, so if the number of wires exceeds 64, continuing to increase the number of wires will increase the cost more rapidly. In contrast, many LiDAR systems use rotating components or rotating mirrors to scan the line 360 degrees around the environment.
Common strategies include deflecting individual transmitters and receivers up or down to increase the laser's field of view. For example, Velodyne's 64-lane LiDAR system has a vertical viewing angle of 26.8 degrees (it has a 360-degree horizontal viewing angle through rotation). This LiDAR can see the top of a 12-meter-high object from 50 meters away.
Corresponding to the different lines of LiDAR, there are different sharpness bands, this is because the data facsimile decreases with distance. Although it is not perfect, higher resolutions can be used for closer objects, because as the distance to the sensor increases, the angle between the transmitters (for example, 2 degrees) will cause the spacing between these dot bands to bigger.
The problems LiDAR needs to overcome
From the vehicle's automatic navigation to LiDAR
- The material of the reflective surface
Since LiDAR is based on the measurement of the time required for the laser pulse to return to the sensor, if the laser hits a highly reflective surface, this will cause problems for the measurement. From a microscopic point of view, most materials have rough surfaces and scatter light in all directions; a small part of this scattered light can always return to the sensor and is sufficient to generate distance data. However, if the surface reflectivity is very high, the light will scatter away from the sensor, then the point cloud in this area cannot be detected, and the data will be incomplete.
- The environment in the air
The environment in the air can also affect the LiDAR readings. Heavy fog and heavy rain will weaken the emitted laser pulse and affect LiDAR. To solve these problems, higher-power lasers have been put into use, but it is not a good solution for smaller, mobile, or power-sensitive application scenarios.
- Data update rate during rotation
Another challenge facing the LiDAR system is the relatively slow update rate when rotating. The update rate of the system is limited by the rotation speed of the complex optics. The fastest rotation rate of the LiDAR system is about 10Hz, which limits the data update rate.
When the sensor rotates, a car traveling at 60 mph travels 8.8 feet in 1/10 second, so the sensor can be said to be incapable of changes that occur within 8.8 feet while the car is passing. Recognizable. More importantly, the coverage of LiDAR (under perfect conditions) is 100-120 meters, which is equivalent to less than 4.5 seconds of driving time for a car traveling at 60 mph.
Perhaps for LiDAR, the high cost of the installation is the biggest challenge it needs to overcome. Although the cost of this technology has been greatly reduced since its application, the cost is still an important obstacle.
The operation method of LiDAR is very straightforward. The principle is the same as that of radar. The difference is that LiDAR uses light waves (infrared), while radar uses electromagnetic waves. Both emit a sequence of pulsed light waves or electromagnetic waves. When the waves travel forward, they will reflect when they encounter objects. When the reflected wave pulse is received and the time of flight is calculated, the distance between each other can be measured. This process is very straightforward mathematical calculations and does not involve algorithms and artificial intelligence. Because the light wave has a short wavelength, it can have accurate resolution and measurement results for the object in front. However, the long wavelength of the radar makes it difficult to distinguish whether the object in front is a pedestrian or a telegraph pole.
LiDAR has been used in other fields before it is applied to automatic navigation. For example, in the field of archaeology, researchers used LiDAR and aerial cameras to discover larger ancient city ruins in Angkor Wat in Cambodia. In addition, LiDAR is also applied to wind turbines to measure the speed and direction of the wind to adjust the best windward angle.
The obstacle to the widespread promotion of LiDAR technology is its high price. LiDAR technology mainly uses a long series of high-power semiconductor lasers and performs a 360-degree rotating scan to obtain a full range of 3D images, whether it is an optical system. Or mechanical rotation is expensive. Recently, to popularize and reduce costs, Lidar has gradually abandoned 360 degrees in its design and replaced it with less than 180 degrees and a shorter detection range, so that surface-emitting lasers (VCSEL) and digital light processor(s) can be used. DLP) or the use of diffractive optical surfaces to generate an array of laser light will be competitive in the cost structure.
In the future, when automatic navigation becomes more and more common, if LiDAR’s infrared laser signals are everywhere, it will also cause concerns about health and safety. Therefore, LiDAR sensing needs to reduce the intensity of the laser light, in other words, it is necessary to reduce the intensity of the laser light. A high-sensitivity photodetector is installed on it.
The general light sensor operates in the small dark current region of the reverse bias voltage of the element. Once a light signal enters, the current will be enhanced, but at least tens of thousands of photons can generate a significant photocurrent. The single-photon light sensor deliberately operates the component in the collapse zone of the reverse bias voltage. If a photon enters, it will cause a substantial collapse, and it can be restored immediately after the collapse. So, if you count the number of crashes of the component, you know how many photons have been collected, so the sensitivity is increased by several orders of magnitude at once.
The invention of science and technology is often not clear at the beginning of its appropriate application scenarios. Just like a single-photon light sensor, it will even be commented as a "solution looking for a problem", but once the time comes, it must come forward. In the future, self-driving cars will achieve level 5 fully autonomous driving. Whether it is a stereo camera lens, LiDAR, radar, or ultrasonic, car manufacturers must have the ability to fusion these different detection systems completely, because there is no single detection system. The measurement system can provide complete data in different weather, distance, and accuracy.