Abstract: This paper studies the real-time detection of lane markings based on machine vision. Based on a discussion of various lane models, a lane model based on segmented switching is proposed. The detection algorithm for lane images is briefly described, and the experimental results are further discussed and analyzed. Experimental results show that the lane recognition algorithm based on the segmented switching lane model can match various road shapes well and has good real-time performance. Keywords: Machine Vision; Lane Detection; Intelligence Steer 0 Introduction In the field of intelligent vehicle research, lane recognition is a key technology for active safety systems such as lane departure warning and lane keeping[1]. An important step in lane recognition technology is the selection of lane models. A suitable lane model can better describe the geometry of the lane and is also conducive to the real-time performance of the detection algorithm. 1 Lane Model 1.1 Segmented Switching Lane Model Many scholars at home and abroad have established a variety of lane models. The simplest lane model is the straight line model. The advantage of this model is that it requires less computation and is conducive to improving the real-time performance of the algorithm. However, the straight line model is not accurate enough for matching curved road sections. In order to make up for this shortcoming of the straight line model, researchers have proposed a variety of curve models, such as the concentric circle curve model, the quadratic curve model, and the approximate spiral curve model. Compared with the straight line model, the curve model can better describe the geometry of the lane line. However, the curve model is more complex and requires more computation in the process of matching with the lane line, which will affect the real-time performance of the algorithm. In order to make up for the shortcomings of the single model, some researchers have proposed the concept of combined models, such as the segmented straight line model [2] and the straight line-parabola model [3]. For the near field of view, the combined model uses straight lines to match the lane line. However, for the selection of the far field of view model, the combined model still faces difficulties. If the straight line model is selected, it does not match well with curved roads. If the curve model is selected, it is not conducive to controlling the amount of computation. To resolve the contradiction between the accuracy of lane model descriptions of actual lane line shapes and the real-time performance of lane recognition algorithms, this paper improves the combined model by designing a segmented switching lane model. Similar to the combined model, this model divides the image into near-field and far-field regions. Since lane lines in the near-field region are essentially straight, the lower half of the segmented switching lane model uses a straight-line model. For the far-field region, based on the road's curvature, the upper half of the segmented switching lane model switches between a straight-line model and a quadratic curve model according to a switching logic. For details, please click: Research on Real-Time Lane Marking Detection Based on Machine Vision