基于定向距离变换耦合多粒子滤波器的车道线检测算法

针对复杂环境下车道线检测精度不高的问题,提出了一种定向距离变换耦合多粒子滤波器的车道线检测算法。首先,利用四点透视映射方法,将输入图像转换为鸟瞰图,使车道边界平行,便于车道检测。引入定向距离变换(oriented distance transform,ODT),将鸟瞰图边缘像素标记到水平和垂直方向上最近的点,寻找初始边界点。其次,利用车道中心、中心到左右边界的角度以及左右车道边界的切角来构建车道线模型,通过分别考虑两个独立的4D粒子空间,以应用于左右车道边界。随后,在车道模型引入多粒子滤波器,利用左右两侧独立传播的粒子来侦测和追踪一对车道边界点,并使用局部线性回归调整得到的边界点。为了优化多粒子滤波器性能,根据粒子状态向量创建动态依赖关系。最后,通过迭代来确定粒子对应的权重,利用多粒子滤波来检测车道线。实验表明,与当前流行车道线检测算法比较,在多种复杂干扰环境下,所提算法具备更高的检测精度与鲁棒性。

Lane line detection based on oriented distance transform coupled multi-particle filter

Aiming at the problem of low accuracy of lane detection in complex environment, a lane detection algorithm based on directional distance transform coupled with multi-particle filter was proposed. Firstly, the four-point perspective mapping method was used to transform the input image into an aerial view, which makes the lane boundary parallel and convenient for lane detection. Oriented distance transform (ODT) was introduced to mark the edge pixels of aerial view to the nearest points in horizontal and vertical directions to find the initial boundary points. Secondly, the lane model was constructed by using the lane center, the angle from the center to the left and the right boundary and the tangent angle of the left and the right lane boundary. Two independent 4D particle spaces were applied to the left and the right lane boundary. Subsequently, a multi-particle filter is introduced into the lane model to detect and track a pair of lane boundary points using particles propagating independently on both sides of the lane, and the boundary points are adjusted by local linear regression. In order to optimize the performance of multi-particle filter, dynamic dependencies were created according to the particle state vector. Finally, the weight of particles is determined by iteration, and the lane line was detected by multi-particle filter. Experiments show that, compared with the current popular lane detection algorithms, the proposed algorithm has higher detection accuracy and robustness in a variety of complex interference environments.