基于神经网络的车道偏移自动检测的研究

车道偏移是交通事故的主要因素之一。通过引入神经网络提出了一种应用机器视觉检测车道偏移的方法。首先通过图像处理从道路图像提取道路标志线。然后通过最小二乘法计算道路图像中道路标志线的参数。以道路图像中道路标志线的参数为输入,车道偏移参数为输出,建立神经网络进行车道偏移检测。实验结果表明误差小于0.5%,方差检验结果也表明人工测量和神经网络的结果无显著差异。新方法将有助于车道偏移自动检测技术的发展。

The study of lane offset detection based on neural network

Lane offset is an important factor of traffic accidents. A method of lane offset detection with machine vision based on neural network is proposed. The lane mark is first obtained from road image via image process. And then the parameters of lane mark in road image are computed out with the method of least squares. There are two parameters of lane mark as input parameters and two parameters of lane offset as output parameters. A neural network of lane offset detection is built. The experiment results show that the error is lower than 0.5%. The analysis of variance by t test also shows that there are no significant difference between the parameters values with manual detection and neural network. So the new method should be helpful for lane offset detection.