Research And Application Of Lane Line Detection Based On Cluster Segmentation

With the increasingly serious traffic safety problem,autonomous driving technology has gradually become a research hotspot.The detection of road information has become particularly important as the basis of autonomous driving research.Vanishing points and lane lines are indispensable components of road information,and they have received more and more attention in related research.The vanishing point provides an important judgment criterion for the automatic driving visual perception system.Existing vanishing point detection methods still cannot meet the real road processing environment in terms of speed and accuracy.This paper proposes a fast vanishing point detection method based on row space features.By analyzing row space features,similar vanishing points in row space are clustered,and then motion vectors are screened for vanishing points in candidate rows.The results show that the average error of the normalized Euclidean distance is 2.37‰ in driving scenes under various lighting conditions.The only candidate line space greatly reduces the computation,making the real-time frame rate 86.Lane line detection provides important information for lane departure warning,keeping assist,and other modules to ensure driving safety.Pixel segmentation is one of the most commonly used deep learning methods for lane line detection.Although deep segmentation has become the mainstream,problems such as slow detection speed and limited receptive field still exist.Aiming at these problems,this paper proposes a lightweight lane line detection algorithm based on learnable cluster segmentation and a self-attention mechanism.The similarity matching is clustered by pixels under row segmentation,and a self-attention mechanism is introduced to reduce the detection range and proportion of background pixels,further significantly reducing the computational cost.The results show that the accuracy rate on the Tu Simple dataset reaches 97.15%,the F1 score of the CULane dataset is 73.5,and the real-time frame rate is 142.7,which solves the problem of slow cluster segmentation and improves the accuracy of row segmentation.In the real vehicle test scene,the algorithm’s misjudgment rate for lane line detection points is only 6.7%.Therefore,the vanishing point and lane line detection algorithm proposed in this paper has extremely fast speed and good generalization performance is suitable for various complex practical scenarios and meets the high standard requirements of autonomous driving technology.