车型自动识别研究
文献类型：学位论文
| 作者 | 李训青 |
| 学位类别 | 工学硕士 |
| 答辩日期 | 2003-07-01 |
| 授予单位 | 中国科学院研究生院 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 刘迎建 |
| 关键词 | 智能交通系统 车牌定位 车型识别 候选区域 Intelligent Transportation System ITS vehicle card localization vehicle type recognition available brand aims |
| 其他题名 | Vehicle Type Recognition Research |
| 学位专业 | 模式识别与智能系统 |
| 中文摘要 | 智能交通系统是2l世纪世界道路交通的发展趋势。在智能交通系统 中，车牌识别和车型分类是为适应快速发展的交通需求应运而生的。 高速公路的不断发展和车辆管理体制的不断完善，使图像场景日益简 单化和标准化，这为以图像理解为基础的智能交通管理系统进入实际应用 领域提供了契机。要识别车辆的厂家和类别（以后简称车型识别），唯有从 图像中的车标入手。首先定位车标，然后对车标加以识别，从而识别出车 辆的类别，这正是本文的思路。 由于车标定位和车型识别可以借鉴的方法较少，所以我们进行了大量 的尝试，提出了一种可行的思路。 首先根据成熟的车牌定位算法找出车牌，根据车牌和车标之间的几何 位置关系确定出车标可能所在区域，以后的所有操作都是针对这一区域进 行的，我们称之为车标区域。然后对车标区域进行图像增强，包括图像的 平滑、去噪、边缘提取、膨胀、二值化等，接着进行连通域分析，得到了 车标的多个候选区域。同时对车标区域进行基于Hausdorff距离的几何形 状定位，找出具有圆形和椭圆的候选区域。然后用学习好的高斯混合模型 和SVM决策面进行精确定位，得到车标的精确位置。最后进行识别，给出 识别结果。 在本实验室收集的车辆图像集上，我们得到了较好的实验结果。 |
| 英文摘要 | Intelligent Transportation System, ITS, is the development trend of the world road traffic in 21st century. In ITS, vehicle card recognition and type classification emerge as the times require of the fast development of traffic. As the developing of speedway and consummating of vehicle management system, the image scenes become simplification and standardization, this gives chances for ITS based of image comprehension to enter the practice application fields. The only way to recognize vehicle's manufacturer and type is to recognize the brand in the vehicle's image. First go to the vehicle brand, then recognize it, which is the sparkle of this thesis. Since nobody does enough work about vehicle brand localization and vehicle type recognition, no method could be used for reference. In practice, we have tried lots of ways and find a perfect and feasible method. Firstly, we find the vehicle card using the perfect vehicle card localization algorithm, make sure the region including brand(brand region) by the geometry location relation between vehicle card and brand. Secondly, manipulate the brand region of the vehicle image, including smoothing, edge detection, inflating, binarization, then connection region analyzing and get several available brand aims. At the same time, use the geometry shape localization algorithm based Hausdorff distance, we get several other brand aims with Circle and Ellipse. Then we use Gauss Mixture Model and SVM to fix on the correct one of these available brand aims. Lastly, we get the recognition result of the vehicle type using k-nearest algorithm. We have achieved inspiring experiment result in the vehicle image test set collected by our laboratory. |
| 语种 | 中文 |
| 其他标识符 | 703 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/6842]  |
| 专题 | 毕业生_硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 李训青. 车型自动识别研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2003. |
入库方式： OAI收割
来源：自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。