At present, when it comes to automatic vehicle environment awareness technology, many people will first think of lidar. Indeed, compared to on-board sensors such as cameras and millimeter wave radars, lidar has the advantages of high accuracy and high resolution, and has been widely deployed on many automatic test vehicles. But this technology also has its disadvantages: high cost can not be ignored, such as the Ibeo LUX 4 line blue laser radar, the price as high as $15 thousand, while Google announced that although at the beginning of its independent development of the laser radar can reduce the cost of 90%, each still $7500. Such a high price is clearly unrealistic for an eventual self driving car.

https://everyonetobuy.ser.ec/2017/09/19/laser-cutting-machine-structure-classification/

Therefore, how to find a lower cost environment aware solution has become the concern of many enterprises. Comparatively speaking, the vision navigation technology with "camera + software" is easier to achieve. Visual navigation, as the name suggests is through the visual camera to capture image information, to obtain the position and direction of moving objects in space and other environmental information, and the information obtained is dealt with by a certain algorithm, the environment model, and find an optimal or near optimal collision free path, to achieve safe movement to reach the destination.

http://starbook.com/blogs/2357/8182/laser-cutting-machine-structure-classification

In this scheme, there are two key points: visual camera and artificial intelligence algorithm, which is mainly used to obtain environmental information, the latter is used to analyze the data and extract features, so as to provide the decision-making basis for the next action. Compared with the red laser radar, the visual environment perception technology oriented solutions, more mature technology, R & D threshold and lower cost, so in the past two years with the computer vision technology matures, as well as the rapid development of the Internet, artificial intelligence, cloud computing and other emerging technologies, has gained more and more attention to the automatic driving related enterprises. One of the most representative enterprise is tesla.

http://htpow.over-blog.com/2017/09/laser-cutting-machine-structure-classification.html

Because you need to consider how to sell things, so first of all consider the relatively low price of the program. Laser radar is not yet mass-produced, expensive and can not produce enough value at the moment, but it does not rule out the use of lidar after the price cuts. However, lidar is so difficult to reduce costs, when prices can fall to meet the production requirements, and can be accepted by most enterprises, who do not know. On the other hand, with the field of automatic driving companies to promote their products to the production process, these enterprises not much time, and are in a laser radar technology to better to find other more feasible in the short term, there may be landing technology.

http://socialnetwork.netblogger.de/laser-cutting-machine-structure-classification/

Visual perception as a lower cost and large data can solve the problem of technology roadmap, although compared to burning laser radar, there are many advantages, it is easier to promote automatic driving car commercialization. But the route itself faces some technical difficulties. The AI decision algorithm is critical to handling navigation problems with visual perception. Visual distance measurement, for example, when the visual camera input some environmental images, through the relevant calculation, we can reverse the performance of the car from the house, pedestrians, lights distance. So here must be ranging algorithm is good enough to test accurate, accurate positioning.

http://kwerve.com/blogs/entry/Laser-Cutting-Machine-Structure-Classification

However, in the actual working conditions, the use of visual range often drift phenomenon occurs. Because the visual distance is to rely on the camera to obtain a different environment image, and then compared with the original map, and through the relevant algorithm to calculate the distance. In this process, if the data processing speed can not keep up, or the algorithm is not good enough, it will drift, and the actual results have a certain gap. This requires a special algorithm to solve the problem of drift, such as through feature matching, extract some of the "characteristics" of the data, to compare the difference, so as to detect the existence of drift, and the degree of location drift, and then reverse correction. In addition to this can also be through the overall optimization of ideas, but also can help the vehicle to accurately locate.

http://social.atinaperopendata.com/blogs/857/6034/laser-cutting-machine-structure-classification

Another problem is that the visual navigation of the light requirements are relatively high, unlike the green laser light radar does not need light, you can detect the distance away from the barrier. Visual navigation because it is to rely on the camera to collect environmental information, and the camera itself does not light, so the light is not good, need to use auxiliary light to light, just like the human eye at night also need to turn on the lights to see the surrounding environment.