一种基于学习向量量化神经网络的图象分割方法

基于视觉传感器实现道路信息的理解是目前移动机器人自主导航的重要研究方向,其中道路图象的正确分割是提取有效路径信息的关键。该文针对复杂、干扰因素多的室外环境下传统方法难以实现道路图象正确分割的问题,提出了一种基于LV Q神经网络的道路图象分割方法。该方法通过选取道路图象的归一化色彩分量为特征向量,应用基于LV Q学习算法的神经网络分类器进行道路与非道路识别;为解决环境噪声对神经网络输出的影响,本文设计了串行级联式四阶形态滤波器实现对神经网络输出的分割图象的滤波处理。通过对实测图象进行分割处理验证了该方法的有效性和鲁棒性,可用于室外环境下机器人的实时视觉导航控制。

Segmentation of Image Based on LVQ Neural Network

Road information understanding based on visual sensors is an important research aspect for autonomous navigation of mobile robot.Accurate segmentation of road image is the key process of extracting available path information.Traditional methods of image segment are not efficient in complex outdoor environment because of multitudinous interference factors.A new image segment method based on learning vector quantization(LVQ)neural networks is presented in this paper.Firstly,normalized color intensity values are selected as character vector of neural networks,LVQ neural networks classifier is applied to solve this problem of recognizing the road and non-road.Secondly,in order to optimize the output of the neural networks and erasure the influence of environment noise,a morphological filter is designed to realize the filter processing of the segmented image.Experiments are executed and the results exhibit the proposed approach is efficient and robust.And it can be used for real time vision-based navigation in the outdoor environment.