In 2015, the industry was still arguing: whether the driver is using a laser radar or a camera. By 2016, things have changed a lot, especially after a car company Autopilot died, the industry is gradually thinking that laser radar is very important. Let's take a look at the related content with the car electronics editor.
Lidar is an indispensable sensor for autonomous driving
The figure below shows the system framework of the unmanned partial algorithm layer. Viewed from the left, this is a sensor input such as a lidar, camera, millimeter wave, GPS, encoder, and IMU. The data of these sensors is input into the system's perceptual algorithm. For this perceptual algorithm, we will process and analyze the data, how to separate the static objects, and how to identify, classify and track the dynamic objects.
Unmanned partial algorithm layer system framework
The acquisition of high-precision maps relies heavily on laser radar and cameras. After acquiring high-precision maps, we combine maps with GPS and IMU, encoders, and real-time sensing environments to perform map matching for positioning. For path planning and motion control, the vehicle is ultimately controlled in conjunction with the vehicle's CAN bus.
In terms of perception, we commonly use laser radar, camera, and millimeter wave radar as sensors for sensing external objects. Among them, the laser radar can be used to identify, classify and track objects, and the camera can also classify and track objects. Millimeter wave radar is mainly used to identify obstacles on objects.
The three sensors of millimeter wave radar, camera and laser radar have some coincidence points. This is determined by the nature of the sensors themselves, and they have their own indispensable features.
The millimeter-wave radar is mainly used for obstacle detection; it is difficult for the camera to obtain the model of the three-dimensional object, including its interference with the environment, and the camera is very good for classifying and tracking the object; the laser radar is generally used. For positioning, obstacle detection, object classification, dynamic object tracking and other applications.
Automatic driving is inseparable from the role of radar
In fact, before 2016, the development of the laser radar industry is still relatively slow. But now, including the urgent needs of upstream suppliers and downstream customers, we strongly hope to promote the cost reduction and mass production of laser radar.
I currently classify lidars into two categories: scanning lidars and non-scanning lidars.
1, scanning laser radar
Mechanical rotary lidar (Laser for transmitting, receiving, coaxial rotation), which is relatively mature at present, is used in downstream unmanned driving. More representative are Velodyen, Ibeo, including the mechanical laser radar that we are currently producing. The hybrid solid state is also a mechanically-rotating laser radar.
The other is MEMS. MEMS-based scanning radars are currently under development, and their principle is to change the optical path through MEMS scanning mirrors.
Another type is phased array laser radar (OPA), which is also a scanning laser radar, because it is realized by point-by-point scanning, that is, changing the optical path of the laser emitting phase between multiple small antennas. .
2, non-scanning laser radar
Flash LiDAR is a scanning laser radar that emits a field of light. For example, Ledder Tech in Switzerland is developing products such as Flash liDAR.
At present, in fact, the current driving force of the industry is quite large, including ourselves, and the main research and development efforts are also placed on solid-state laser radar. I believe that soon everyone can see this low-cost laser radar, from research and development, sample to commercial, may be faster than originally expected. Because not a laser radar company is working hard, but the entire industry chain is working hard.
The point cloud is a massive point cloud data that expresses the spatial distribution of the target and the characteristics of the target surface under the same space advocacy system.
The point cloud data generating device can be generated by a laser radar or a depth camera. According to the point cloud obtained by the laser radar, including three-dimensional coordinates (XYZ) and laser reflection intensity (Intensity).
Application of Lidar in autonomous driving: positioning
The most important part of the application of laser radar in autonomous driving is positioning: the position is determined, and the unmanned vehicle knows where to go and how to go. So, determining "where am I" is the first and very important step.
Now use RTK, differential GPS, and also do it with lidar or vision. However, RTK is still subject to signal interference. Especially in some cities, buildings and trees, as well as tunnels and tunnels, its signal is easily interrupted.
Another is based on visual positioning. It is actually related to the characteristics of his own vision and has a strong dependence on the environment. For example, in the case of backlight or rain and snow, this positioning is prone to failure.
Automatic driving is inseparable from the role of radar
Lidar positioning, we get a predicted global position through IMU, inertial navigation system, encoder and GPS. When the lidar scans a single point cloud data in real time, we will combine the single point cloud data for matching and feature extraction.
These features include features of the surrounding point line surfaces such as roadsides, lane lines, and the like. For high-precision maps, the extracted features are matched with the features extracted in real time, and finally the accurate vehicle body speed is obtained, which is the positioning process of the laser radar.
The second application of lidar: the detection and classification of obstacles.
For obstacle detection and classification, there are currently applications of vision and lidar, which are not in conflict. Lidar does not rely on illumination, its viewing angle is 360 degrees, the amount of calculation is relatively small, and it can be scanned in real time. Currently, it is generally used within 100 milliseconds. In the process of scanning, the laser radar first identifies obstacles, knows the position of the obstacle in space, and then classifies according to the existing obstacles.
As shown above, we first get an original point cloud data, extract the ground points, and perform obstacle segmentation on non-ground points.
For example, cars and people, we divide these obstacles into independent individuals, and then separate and separate the individual to match, thus classifying the obstacles and tracking the objects.
To put it simply, we first remove the ground points, get obstacles, segment the obstacles, and then segment the obstacles for classification tracking.
The process of tracking, first of all, is to divide the point cloud and use the point cloud to make the associated target. We know whether the last one and the next one belong to the same object, then perform target tracking and output target tracking information.
Application status of laser radar
We started to supply in April this year, and we are also aware of the current status of the application of laser radar in the industry.
The first is the lack of sensors. On the one hand, the current laser radar is more expensive, the delivery cycle is also very long, and there are not many companies that can generally use laser radar. The lack of sensors, the immaturity of the solution, the accumulation of talents in the point cloud algorithm is too small, and the lidar cannot exert its maximum power.
For the driverless team, in addition to the lidar algorithm of the lidar, they may also do camera algorithm, millimeter wave radar algorithm and multi-sensor fusion, including positioning, path planning, decision control and car change. They did a lot of technical points and couldn't focus, which led them to be forced to stretch the front.
So in April of this year, we proposed the Purmisius plan for the Lidar solution. We hope that the essence of this plan is an attitude of open sharing, accelerating the commercialization of the entire driverless.
Automatic driving is inseparable from the role of radar
The entire Prometheus module is based on what the laser radar can do, including: positioning, lane detection, roadside detection, obstacle recognition, obstacle classification and tracking algorithm modules.
Whether it's a low-speed car, a park car or a logistics car or a car driving on a highway, we all hope that we can contribute.
The above figure is based on lidar lane detection and roadside detection. The lid line detection based on lidar is quite good. The lane lines on the road are generally white and yellow, so we will first make the reflection intensity in advance, so that the laser radar can easily extract the lane line out. Because of disturbances during the day and night.
The roadside detection can do some training according to the geometry of the roadside. Now the road can be extracted at a height of more than ten centimeters.
We can look at the half of the picture: red indicates the extracted lane line, light color is the extracted road edge, the middle is the dotted line, and the two sides are the solid line. This accuracy is good, including the left and right turn of the road. If there are multiple frame iterations in the future, the effect will be better.
Lidar tracking of objects is equivalent to calculating each object identified, and whether it is a bicycle, truck, pedestrian or private car.
After the identification, in addition to knowing the speed of the self-driving body car, it is also possible to track the speed of the preceding vehicle and the lateral and longitudinal distance of the preceding vehicle from the vehicle. The laser radar output is not the original point cloud data, but also the location and classification of each obstacle space, which type, and speed information.
There are many things that laser radar can do, including positioning, obstacle detection, classification and tracking, lane line detection, and edge detection. In the work of perception, laser radar plays a very important role.
The above is about the role of auto-driving in the auto-driving without radar. If you want to know more information, please pay more attention to it. E-engineering will provide you with more complete, more detailed and updated information.
CD Car Phone Holder,Car CD Slot Mount Holder,Car Phone Mount Holder,Car Accessory CD Phone Holder
Ningbo Luke Automotive Supplies Ltd. , https://www.nbluke.com