基于LDA与混沌粒子群算法的车道线识别方法研究

Lane detection by LDA and chaos particle swarm optimization

与传统的车道线检测算法不同,本文采用LDA算法对道路图像进行针对性灰度化处理。加大车道线与道路的差异,然后使用抛物线模型对车道线进行拟合,采用混沌粒子群算法对抛物线参数进行优化,以车道线的灰度特征和梯度特征作为混沌粒子群的适应度函数,经过多次的迭代得到抛物线拟合车道线的参数最优值,进而识别出车道线。实验结果表明,本文算法能在复杂环境下识别出车道线,对视频帧序列中的车道线连续追踪具有良好效果。

The method proposed in this paper is different from the traditional road lane detection algorithm. Firstly, LDA（Linear Discriminant Analysis） is used to convert road images to gray images directly, increasing the difference between the lane and the road. Secondly, the road lane is fitted by parabola model, and CPSO（Chaos Particle Swarm Optimization） is applied to optimize the parameters of parabolic model. The CPSO fitness function consists of features of grayscale and gradient. After several iterations, the optimal parameter value of parabola can be obtained, and then the road lane in the image can be detected out. Experiment results show that the proposed algorithm could detect road lane in complex environment. Also, it has a good performance on road lane tracking in the video frame sequence.