How do autonomous driving environmental sensors detect the type of lane markings? What is the logic

Hello, lane markings detection is essentially a parameter estimation problem. When doing practical engineering, there are two questions: 1. What should be chosen for the mathematical model of lane markings? 2. What type of lane markings are detected?

For the first problem, the commonly used lane line models are divided into two categories: parametric models such as line parabola cubic, poly-line spline, and clothoid, which can describe the entire lane line shape with only a few parameters; There are also data-based, such as support vector regression, gaussian process regression, which need to be supported by the correct data and learn the corresponding parameters.

The problem of parameter estimation for lane markings is not simple, because the data itself has outliers in addition to noise. An outerrod can invalidate the traditional method of least squares. Hough Transform, Ransac, Least Trimmed Square, Bayesian Filter can all be used for robust parameter estimation.

For the second question, the detected lane lines can be divided into solid white lines, solid yellow lines, etc., to use the classification algorithm, I don't understand very well, so I won't force the answer.

PS: The use of cameras to detect lane markings is already standard for ADAS, but, but, but, for unmanned driving, the sensitivity of the camera to the environment causes lane marking detection to sometimes fail. At this time, it is best to use a laser sensor as a supplement.

PPS: Taking it a step further, why do you have to inspect lane markings? Stanford's Ph.D. didn't involve lane markings, they stitched the effective information on the ground (including lane markings, pedestrian crossings, steering arrows, and even cracks) into a high-precision map, and the positioning was OK.

1. Sensor technology: Self-driving cars are equipped with various sensors to understand the surrounding environment, roads, and traffic conditions. Optical cameras include single cameras, multiple cameras, Doppler radar includes short-range radar, long-range radar, and lidar is the rotating machine on the roof of the car, GPS positioning device, etc., which constitute the eyes of the car to understand the environment.

2. Chip technology: that is, it can process the data collected by multiple sensors and integrate a super chip similar to a small computer, which greatly reduces the volume and cost of the "total computer" of the car, and can be applied to the car. Otherwise, there will be no place for people to sit in the car, and ordinary people will not be able to afford cars with these huge computer clusters.

3. Operating system: The computer control system combines the processing results with the operating hardware to realize acceleration and deceleration, braking and parking, changing direction and avoidance, and man-machine dialogue. Driverless cars have the ability to replace human manoeuvres.

4. Network technology: In order to be able to go on the road, driverless cars must have the function of contact and identification with the Internet and local area network, including car-to-car contact dialogue, car and satellite communication, car and weather forecast contact, car and traffic command network contact, in order to correctly identify and choose the road, correctly obey the command of the traffic police, correctly decide to pass through the intersection, correctly avoid danger and drive safely. In short, the Long March has just taken its first step, and it is still quite far away from entering the people's homes.

For example, the thing that always rotates on the roof of the car makes people feel very **. Automakers are really engaged in unmanned driving, and they themselves have pushed themselves to chip companies, computer companies, and network companies, and become a member of the assembly plant, which is also very reluctant.