Vision is the most important functional means of human cognition. Biological research shows that 75% of humans access external information rely on the visual system, and in the driving environment this proportion is even as high as 90%. If the human visual system can be applied to the field of autonomous driving, it will undoubtedly greatly improve the accuracy of autonomous driving, which is one of the most popular research directions in computer science and autonomous driving. It is machine vision technology. The development of machine vision technology has been more than 20 years old, and the real revolutionary progress is the introduction of Moore's visual computing theory. It is possible to realize the same functions of the human visual system by implementing neural network related algorithms. . In general, the machine vision system includes a lens, a camera system, and an image processing system, and the core is a dedicated high-speed image processing unit, which compares the large amount of digitized information stored with the template library information, and quickly obtains In conclusion, its speed and accuracy are key indicators. This is mainly achieved by efficient and reasonable algorithms and powerful chips. At present, there are a variety of high-efficiency vision-specific hardware processors and chips and other electronic devices on the market, and with the advancement of computer technology, more advanced algorithms have been successively invented, such as the use of grid distributed processing systems can effectively improve the operation. effectiveness. The core issue of machine vision in the future will be an in-depth understanding of the image. There are two main aspects to the application of machine vision in autonomous driving: Obstacle detection The accuracy of obstacle detection is an important guarantee for the safety of the vehicle during automatic driving. During the driving process, the occurrence of obstacles is unpredictable, and it is impossible to avoid obstacles according to the existing electronic map. It can only be found and processed in time while the vehicle is running. Currently, due to the immaturity of the autonomous driving environment, there is no uniform standard for the definition of obstacles. Therefore, it can be considered that all objects that may hinder the normal running of the vehicle and the abnormal terrain that affects the passage of the vehicle are obstacles during the running of the vehicle. At present, there are three main obstacle detection algorithms: 1. feature-based obstacle detection; 2. obstacle detection based on optical flow field; 3. obstacle detection based on stereo vision. Among the three algorithms, obstacle detection based on stereo vision has become a mainstream research direction because it does not require a priori knowledge of obstacles, whether or not the obstacles move or not, and directly obtains the actual position of the obstacles. However, it has higher requirements for camera calibration. While the vehicle is running, the camera calibration parameters will drift and the camera needs to be dynamically calibrated. Road detection Automatic navigation is a necessary condition for automatic driving. During automatic driving, road detection is mainly to determine the position and direction of the vehicle in the road in order to control the vehicle to follow the correct route. In addition, it determines the search range for subsequent obstacle detection, and narrows the search space for obstacle detection, reducing algorithm complexity and misrecognition rate. However, due to the variety of roads in reality, road detection is a very complicated problem due to the influence of various environmental factors such as illumination and climate. There is still no universal algorithm, and the existing algorithms basically make certain assumptions about the road. The commonly used assumptions are: 1 specific interest area assumptions; 2 road equal width assumptions; 3 road flat assumptions. In addition, road flat assumptions also provide a reference for obstacle definition. At present, machine vision technology has not been applied in large-scale in automatic driving. In fact, this is not a hardware problem. In fact, the application of camera technology in automobiles is very mature, such as the driving recorder of Shanling Technology, wide-angle field of view. The functions of reversing images and the like are all fully equipped, and the chip technology has been able to efficiently perform image compression processing. The final difficulty lies in the simulation of the neural network. Ideal for indoor and outdoor installations, including street lighting and architectural illumination, these highly efficient LED Drivers are protected from dust and moisture, providing outputs up to 240W.
Top features include:
100-347V Waterproof LED Driver Street Light 347V LED Driver,100-347V LED Driver,AC347V LED Waterproof Driver,Waterproof LED Driver,wide input AC, cUL listed Driver ShenZhen Fahold Electronic Limited , https://www.fahold.com
October 31, 2020