-
自主式地面车辆 ALV（Autonomous Land Vehicle）也称智能车，是室外轮式机器人在交通领域的重要应用。它使用车载的视觉，激光雷达，超声测距仪，全球定位系统等传感器来感知环境，完成自主行驶。因此，可靠的道路环境感知技术成为了自主式地面车辆的基础。 道路环境感知的目的在于对自然场景中的道路周边地表覆盖物进行准确分类。本课题利用 Outex 大学的地表覆盖物纹理数据库以及大学创建的越野地表覆盖物数据库作为数据源，抽取图像的灰度共生矩阵（GLCM）,局部二进制模式(LBP),Gabor 特征以及颜色特征构成特征空间，并在此空间上重点研究了用KSVD 学习类内字典和构建带标签全局字典（LCKSVD）的稀疏表示分类方法。利用Matlab 数值计算工具对数据集进行测试实验，实验表明颜色特征在地表覆盖物分类中比较重要，而标签一致字典学习算法对于低维样本效果可以达到 97%以上的分类正确率。 59569
- 上一篇：B5H5的结构和性能研究
- 下一篇：不同能量毫秒激光与铝相互作用的温度计算
毕业论文关键词 ALV 地表分类 稀疏表示 KSVD
Title Some Research on Land Cover Classification Method in Cross-country Environment
Abstract Autonomous Land Vehicle,which is also called Smart Car,is an important application of wheeled robots in the transport sector.All kinds of sensors such as camera,laser radar and GPS have been used to help the robot sensing the environment around，the robot then can drive automaticlly.Therefore,a robust road environment sensing technology is the foundation of ALV. The purpos of environment sensing is to classify all kinds of land cover correctly from nature scene. Our project takes the Outex texture database and the land cover database which is created by Nanjing University of Sciene and Technology as the data source. We then extract the color ,LBP ,GLCM ,Gabor feature from images to build feature space. Two methods for dictionary learning in the model of sparse representation are reseached based on the feature space. The first method is to learn a dictionary for each class with KSVD. Learning a label consistent dictionary by KSVD framework is the second way. Some experiments for land cover classification on the dataset have been done, the results show that color feature contributes more to the classification of land cover and the classification accuracy of Label consistent dictionary learning method can up to 97% when the feature space has low dimension.
Keywords ALV Land Cover Classification Sparse Representation KSVD
目次
1引言1
1.1越野行车环境中地表覆盖物分类问题综述1
1.2基于稀疏表示模型的地表分类问题框架3
2图像的常用特征提取算法5
2.1颜色矩（ColorMoments）特征5
2.2局部二进制模式(LBP)特征6
2.3灰度共生矩阵(GLCM)7
2.4Gabor纹理特征9
2.5特征分类性能的比较10
3稀疏表示模型中的字典学习方法13
3.1稀疏模型用于分类的基础13
3.2直接利用样本构建字典用于稀疏分类(N_SRC)15
3.3KSVD学习字典用于稀疏分类（KSVD_SRC）16
3.4标签一致性KSVD字典学习方法用于分类(LCKSVD_SRC)17
4实验结果21
4.1Njust_64数据库上的单特征分类对比实验21
4.2所有数据库上的全特征分类对比实验21
结论23
致谢24
参考文献25
1 引言 本课题来源于自主车项目的环境感知子课题，本章将首先对越野行车环境中地表覆盖物分类问题进行综述，继而提出本课题的解决思路和系统框架。