As of now, the latest news shows that Uber has reduced the number of green laser pointer radars in driverless test vehicles from five to only one on the roof, resulting in more blind spots in vehicles and the inability to fully detect pedestrians. Is the main cause of the accident. Its lidar supplier Velodyne also believes that radar must be installed on both sides of the body to avoid hitting pedestrians. Although it is said that Velodyne’s motivation is to clear the relationship as much as possible, the importance of lidar in autonomous driving is self-evident.
Compared with sensors such as cameras, in addition to being able to generate three-dimensional position models, lidars have longer detection distances, higher measurement accuracy, and more sensitive response speed, and are not affected by ambient light. In addition, the data fed back by lidar is easier to read and analyze by the computer, so it has become one of the core technologies of autonomous driving. However, lidar is also easily affected by weather, especially rain and fog weather will scatter the laser, which will affect the accuracy of lidar and the problem of inaccurate detection distance.
In order to solve this problem, MIT has developed a new imaging system that can accurately calculate the distance by using lasers in dense fog. As we all know, lidar mainly calculates the distance to the object in front by the short-pulse laser emitted and the reflection time reflected from the object in front. Due to the dense fog scattering phenomenon, the time when the laser beam is reflected back is often different from the sunny day, which leads to the problem of inaccurate detection by the green laser pointer radar. MIT researchers analyzed and found that no matter how thick the fog may be, the reflection time of the scattered laser light always follows a very specific distribution pattern. When the camera calculates the number of photons returned to the sensor at a frequency of one-millionth of a second and plots these results as a graph, the system can apply specific mathematical filters to find data spikes and thus be accurately hidden in the fog The distance of the actual object.