基于支持向量机的激光雷达可通行区域检测  

Detection of accessible area by Laser radar based on support vector machine

作　　者：任国亭 王新晴[1] 黄杰[1] 孟凡杰[1] 

机构地区：[1]解放军理工大学野战工程学院,南京210007

出　　处：《计算机应用》2017年第A02期82-84,106,共4页journal of Computer Applications

基　　金：国家重点研发计划项目(2016YFC0802900);国家自然科学基金资助项目(61671470)

摘　　要：针对激光雷达识别环境道路,常采用的基于高度差方法具有单一特征阈值选取困难、识别准确率较低的问题,提出了一种基于支持向量机(SVM)的可通行区域检测方法。首先根据等效密度法划分扇块,对扇块中的数据点进行特征提取,采用多种特征融合对扇块进行描述,使用支持向量机对样本特征进行训练,之后通过训练模型对激光雷达数据进行检测。激光雷达采集到的数据为离散分布的数据点,而且分布不均匀,进行扇块划分时会有很多没有数据点的扇块区域,提出基于邻域扇块属性的区域扩展方法对空白扇块进一步处理,最终得到可通行的道路区域。对于校园道路,传统的基于高度差方法检测道路准确率能够达到90%左右,而提出的检测方法针对校园三种路况道路可通行区域准确率能够分别为98.3%、97.8%以及97.9%,能够有效检测道路可通行区域,对于无人车路径规划以及安全行驶能够提供比较准确的信息。Concerning the problem that the height difference method has difficulty of single feature threshold selection and low recognition accuracy, a new method based on Support Vector Machine（ SVM） was proposed. Firstly, according to the equivalent density method, the Lidar（ Laser radar） data was divided into fan blocks and features of data points in blocks were extracted. Then, the fan blocks were described by multi-feature fusion, and the model of SVM was trained on the sample features. Finally, the Lidar data was detected by the model. The Lidar data is a collection of discrete data points, which are not uniform, so there will be a lot of fan partitions without data points. In this paper, the method of region expansion based on the property of neighborhood block was proposed to deal with the blank block further. For the campus road, the detection accuracy of the traditional height difference detection method can reach about 90%, and the detection accuracies of the proposed method in this paper is 98. 3%, 97. 8% and 97. 9% respectively for three types of road traffic detection areas. This method can effectively detect the road area, and provide more accurate information for path planning and safe driving of unmanned vehicles.

关 键 词：激光雷达 等效密度 支持向量机 特征融合 区域扩展 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]