How does the automatic driving environment sensor detect the lane line type? What is the logic of judgment?

Lane detection is essentially a parameter estimation problem. There are two questions to answer when doing practical engineering: 1, what is the mathematical model of lane line? 2. What kind of lane lines are detected?

For the first question, the commonly used lane line models are divided into two categories. Parametric models, such as straight line/parabola/cubic curve, polyline/spline curve and cyclotron, only need a few parameters to describe the shape of the whole lane line. There are also data-based methods, such as support vector regression and Gaussian process regression. This method needs correct data support and learning corresponding parameters.

The problem of parameter estimation of lane lines is not simple, because the data itself includes outliers besides noise. Abnormal values will invalidate the traditional least square method. Hough transform, ransac, minimum trimmed square sum Bayesian filter can all be used for robust parameter estimation.

For the second question, the detected lane lines can be divided into white solid lines, yellow solid lines and so on. I don't understand the classification algorithm very well, so I don't answer it.

It is a standard configuration for ADAS to use a camera to detect lane lines. However, for unmanned drivers, the sensitivity of the camera to the environment sometimes leads to lane line detection failure. At this time, it is best to use a laser sensor as a supplement.

Pps: Further, why should we check the lane lines? Stanford's doctoral thesis did not cover lane lines. They spliced the effective information on the ground (including lane lines, crosswalks, turning arrows and even cracks) into high-precision maps and located them online.