摘要
为实现无人装备在野外复杂环境下,对非结构化道路进行精准高效识别,提出了一种野外场景自适应道路识别算法。该算法加入新的自适应预处理算法,对野外环境图像进行类别划分,对不同类型图像进行针对性预处理;将图像从像素级划分成区域级,划分为同质的超像素块,通过提取、融合超像素块区分度高的颜色、纹理、位置、形状特征,构造新的具有更高区分度的超像素块综合特征向量;通过动态选取道路标识样本,改进了传统拉普拉斯支持向量机算法,利用改进后的算法学习和训练出了超像素块分类器,成功实现了野外复杂场景下道路的精准高效识别。指定数据库检测结果表明,道路识别精度达91.9%,具有较高的道路检测精度和较好的实时性,能够实现对野外非结构化道路精准有效识别。
In order to realize the accurate and efficient identification of unstructured roads under the complex environment of unmanned equipment, this paper proposes an adaptive algorithm for field recognition in the field. The algorithm adds a new adaptive preprocessing algorithm to classify the field environment images and perform targeted preprocessing on different types of images. The image is divided from the pixel level to the regional level and divided into homogenous super pixel blocks. It combines the color,texture,position,and shape features with high discrimination in superpixel blocks,constructs a new superpixel block synthesis feature vector with higher discrimination,and improves the traditional Laplace support vector by dynamically selecting road identity samples. Machine algorithm,using the improved algorithm to learn and train the super pixel block classifier,successfully realized accurate and efficient recognition of roads in complex field scenes. The specified database detection results show that the road recognition accuracy is 91. 9%, and it has higher road detection accuracy and better real-time performance,and can accurately and effectively identify unstructured roads in the field.
作者
华夏
王新晴
俞垚魏
孟凡杰
马昭烨
王东
邵发明
HUA Xia;WANG Xinqing;YU Yaowei;MENG Fanjie;MA Shaoye;WANG Dong;SHAO Faming(PLA Army Engineering University, Nanjing 210018, China)
出处
《兵器装备工程学报》
CAS
北大核心
2018年第6期165-170,共6页
Journal of Ordnance Equipment Engineering
关键词
非结构化道路
自适应
多特征
超像素
支持向量机
机器自学习
unstructured road
adaptive
multi feature
super pixel
support vector machine
machine self learning