一种基于几何约束的轨道提取方法研究
发布时间：2019-07-15 18:48
【摘要】：针对基于Hough变换或螺旋曲线模型的视觉轨道检测方法存在的不足,本文提出了一种基于几何约束的轨道提取方法.该方法利用摄像机和轨道平面之间的成像关系近似满足单因矩阵的特点,利用逆透视映射(IPM)将输入图像转换为Bird-view图像,并采用一种改进的边缘检测方法进行边缘检测.然后将二值化的边缘图像在垂直方向上分割为多个区段,在每个区段上,利用先验知识生成的系列模板图像,对分段IPM图像进行去噪处理和Chamfer距离变换后进行距离匹配检测,将轨道检测转换为一个二维匹配搜索过程.在分段检测结果的基础上,进一步利用曲线拟合得到边缘图像中完整的轨道曲线方程.该曲线方程通过已知的单因矩阵转换为原始图像中的曲线描述,实现在原始图像中的检测和定位.实验验证了所提方法的可行性和可靠性.
[Abstract]:Aiming at the shortcomings of visual orbit detection method based on Hough transform or spiral curve model, a trajectory extraction method based on geometric constraints is proposed in this paper. In this method, the imaging relationship between camera and orbit plane is approximately satisfied with the characteristics of single factor matrix, the input image is converted into Bird-view image by inverse perspective mapping (IPM), and an improved edge detection method is used for edge detection. Then the binary edge image is divided into several sections in the vertical direction. On each section, the segmented IPM image is de-noised and the distance matching is detected after Chamfer distance transformation by using a series of template images generated by prior knowledge, and the orbit detection is transformed into a two-dimensional matching search process. Based on the piecewise detection results, the complete orbital curve equation in the edge image is obtained by curve fitting. The curve equation is transformed into the curve description in the original image by the known single factor matrix, and the detection and location in the original image are realized. The feasibility and reliability of the proposed method are verified by experiments.
【作者单位】： 北京交通大学电子信息工程学院;
【基金】：国家自然科学基金 国家科技支撑计划项目 中央高校基本科研业务费专项资金~~
【分类号】：TP391.41;U216.3
本文编号：2514814
[Abstract]:Aiming at the shortcomings of visual orbit detection method based on Hough transform or spiral curve model, a trajectory extraction method based on geometric constraints is proposed in this paper. In this method, the imaging relationship between camera and orbit plane is approximately satisfied with the characteristics of single factor matrix, the input image is converted into Bird-view image by inverse perspective mapping (IPM), and an improved edge detection method is used for edge detection. Then the binary edge image is divided into several sections in the vertical direction. On each section, the segmented IPM image is de-noised and the distance matching is detected after Chamfer distance transformation by using a series of template images generated by prior knowledge, and the orbit detection is transformed into a two-dimensional matching search process. Based on the piecewise detection results, the complete orbital curve equation in the edge image is obtained by curve fitting. The curve equation is transformed into the curve description in the original image by the known single factor matrix, and the detection and location in the original image are realized. The feasibility and reliability of the proposed method are verified by experiments.
【作者单位】： 北京交通大学电子信息工程学院;
【基金】：国家自然科学基金 国家科技支撑计划项目 中央高校基本科研业务费专项资金~~
【分类号】：TP391.41;U216.3
【相似文献】
相关期刊论文 前1条
1 严隽耄;;具有任意轮廓形状的轮轨空间几何约束的研究[J];西南交通大学学报;1983年03期
，本文编号：2514814
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2514814.html
