Analysis of sensor technology in automatic driving

In recent years, with the continuous expansion and deepening of the application field of artificial intelligence technology, automatic driving has gradually entered the public's field of vision, driving in the eyes of consumers has become a light thing. In the development of autonomous driving technology, the choice of laser radar or camera as the main sensor is the primary problem, they represent two completely different systems - the laser SLAM (Simultaneous Localization And Mapping) and the visual SLAM.

1, the camera

Phil Magney, founder and head of VSI Labs, once said that perception is a major area of the autonomous car software stack, and there are many innovations in it, and the camera is an essential sensor. On the evening of March 13, 2017, Intel announced the acquisition of Mobileye for $15.3 billion. Why did Intel achieve the biggest transaction in the auto industry last year at a premium of one-third?

In 1999, two researchers from the Hebrew University of Israel founded Mobileye, a company dedicated to the vision system of cameras and software algorithms for solving vehicle driving problems. It is precisely because vision is the only indispensable sensor technology in autonomous vehicles, and based on Mobileye's leading position in autonomous driving car vision, Intel has hardly contributed to this marriage. Intel said it is combining Mobileye's "computer vision, sensing, convergence, map construction and drive strategy" with Intel's "open computing platform."

It is understood that the in-vehicle camera is the basis for implementing many kinds of early warning and recognition type ADAS (Advanced Driver Assistance System) functions. Among many ADAS functions, the visual image system is relatively basic, and the driver is more intuitive, and the camera is visual image processing. The basis of the system.

Many functions such as lane departure warning, forward collision warning, traffic sign recognition, pedestrian collision warning, driver fatigue warning, etc. can be realized by means of the camera, and even some functions can only be realized by the camera. In addition, the camera can not only capture images with high resolution, but also better classify objects, so what are their disadvantages? "The data depth of the camera is not as good as the lidar," Magney said.

2, laser radar

With the fiery heat of autonomous driving, the laser radar has been sought after as never before. Lidar (Light DetecTIon And Ranging) is a combination of laser, GPS and inertial measurement devices. It is able to distinguish pedestrian and character posters in real movement, model in three-dimensional space, detect static objects, and accurately measure distance. Its working principle is also well understood. The radar system that detects the target position and velocity by emitting a laser beam has the advantages of high measurement accuracy and direction.

In the ADAS system, the laser radar obtains characteristic data such as the position and moving speed of the target object based on the TOF flight time principle through the lens, laser transmitting and receiving device and transmits it to the data processor; at the same time, the speed, acceleration, direction, etc. of the car The feature data will also be transferred to the data processor via the CAN bus.

Subsequently, the data processor comprehensively processes the information data of the target object and the vehicle itself and issues a corresponding passive warning command or active control command according to the processing result, thereby implementing the assisted driving function. The current mainstream laser radar in the market or in autonomous driving projects accounts for about 90% of the applications of autonomous driving. Laser-radars are basically used in driverless cars developed by companies such as Google, Audi, Ford and Baidu. In addition, according to the number of wire harnesses, laser radar can be divided into single-line laser radar and multi-line laser radar. Single-line lasers are mainly used to avoid obstacles. They are accurate in testing the distance and accuracy of obstacles around them. However, single-wire beams can only be scanned in a plane and cannot measure the height of objects. Multi-line laser radar has become a deficiency for single-line laser radar. , there is a qualitative change in dimension elevation and scene restoration, which can identify the height information of the object.

It is understood that the current internationally launched four-wire, 8-wire, 16-wire, 32-wire and 64-wire, multi-line laser radar is mainly used in radar imaging of automobiles.

Unmanned sensors used for control are either cameras or lidars. This is a consensus that has been reached in the industry, but which one is used is still in dispute. According to some views, there is not such a contradiction between the two. The two technologies have their own advantages and disadvantages. Complementing each other is a relatively achievable route at this stage.

From a deeper perspective, the choice of sensors is actually a matter of automatic driving route selection. If autonomous driving is to be anthropomorphized, then Lidar is basically not needed. The vehicle relies on its own knowledge base and rule base to complete the dynamic judgment of the surrounding environment and vehicle control.

Some people believe that the problem between the two is which side of the real industry development is more focused. In the case of Tesla, it focuses on the camera and the millimeter wave sensor. However, due to the frequent occurrence of Tesla accidents, more and more enterprises are beginning to question the relatively low cost and easy to promote. In addition, the media said that Wisla began testing the model series equipped with lidar in North America. Autopilot is a systematic project. Sensors are indispensable and can't work independently. Based on the perfect combination of reality, the autopilot can go as fast as possible.