Tesla autopilot upgrade technical route from the camera officially turned to the radar 9c8950

Upgrade of automatic driving Tesla: technical route from the camera to the official radar NVIDIA today, Tesla officially to the owners push air upgrade version 8, the most notable is the Autopilot automatic auxiliary drive upgrade. After the upgrade, Tesla Autopilot will officially change the technical route, previously, Tesla more rely on the main camera and image processing system, and the radar is the auxiliary sensors, and the future of Tesla automatic driving auxiliary function will bear more by radar. The radar belongs to Autopilot auxiliary driving supporting hardware, equipped with the October after the factory on the Tesla models in 2014. The technical route changes, characteristics of radar, Tesla respectively and related coping methods to explain: first wavelength photon radar will advantages through fog, dust, rain, snow and so on, which makes the automatic driving in bad weather and more safety. In the past, the camera technology route is confirmed by optical visual image, in fact, the limitations of the camera can be understood as the limitations of the human eye, it is always necessary to ensure that the front line of sight through the wiper and other items. And the radar does not need this, which is why Google driverless car can get rid of wiper. Secondly, metal objects, human, wood products and plastic products painted on radar, either directly or direct reflection, transmission, and this is a problem, because the function of radar must ensure to identify obstacles. The metal items with a concave convex surface even because the reflected signal amplification, the vehicle of the metal goods volume error estimation. For example, a throw in a roadside can, if the concave bottom toward the front, will be shown as a dangerous obstacle of large objects, if it leads to the owner on the brakes to avoid cans, which constitutes a false alarm. How to solve this part of the problem, Tesla gives the following solutions: first, in order to solve the false recognition of obstacles resulting in the first part of the frequent braking problem, Tesla’s solution is to create a more detailed data point cloud. 8 air upgrades to existing radar system functions were unlocked, so that each radar can detect than the surrounding objects before six times, and has the ability to obtain more detailed information on each object. Point cloud is more abundant, the more accurate identification of obstacles. Secondly, in order to reduce the false identification, Tesla used a 1/10 second speed integrated radar image frame, compiled into a realistic scene of 3D" image ". Generally speaking, through a single image frame, it is difficult to know whether the mobile object, static, or just a reflection caused by the virtual image. By comparing the continuous image frames containing the vehicle speed and the desired trajectory information, the vehicle can distinguish whether the object is real or not, and evaluate the probability of collision. The third part is more complicated. When the upstream highway running vehicle is approaching a sign at the top of slope, or a road located in the downstream bridge under the bridge, the vision of the head, often looks like a collision course. The navigation and high precision GPS is not enough to determine whether the vehicle will be through this section to the vehicle close;相关的主题文章: