【作者】 叶艳芳；

【导师】 黄席樾；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 在智能车辆的信息感知中,对前方行驶车辆的检测是其核心任务和重要前提之一。本文以智能主动安全预警系统为背景,以城区复杂背景条件下行驶的车辆为研究对象,对小波变换在目标检测领域中的应用进行了讨论和研究,并基于以上理论提出了一种基于单目视觉的车辆检测的新方法。在车辆检测算法的研究方面,首先用三层金字塔高斯滤波器对原始图像进行平滑,然后采用一阶导数算子——3×3的Sobel算子提取平滑后的图像中包含的边缘信息。在得到图像全局轮廓的基础上,根据投影映射原理和统计数据求得图像中车宽阈值,运用“投票”机制来求取图像中存在的轮廓对称轴,并将对称轴位置映射到一个一维向量中。然后利用小波模极大值原理,通过小波变换检测出一维投票结果向量中的信号突变点,在整个像平面的对称轴集合中探测出候选车辆对称轴。得到候选车辆对称轴后,为了产生可用于确认判别的车辆外接矩阵Region of Interests(ROI),采用了边缘特征作为车辆左右边界定位的依据,并根据底部阴影特征来定位下边界,而对定位准确度要求较低的上边界则简单采用外形高宽比来检测。最后,通过基于局部信息熵和局部灰度对称性测度的假设验证对ROI区域进行验证并对验证后的车辆区域进行滤波,得到最终检测结果。大量的实验数据表明,本文算法具有较高的检测精度,较高的算法稳定性和运行鲁棒性,能较好地满足实际应用的需求。同时,由于本文算法是在全局图像中进行车辆检测,并未采用车道线等通常的约束条件,因此本文算法可良好地应用于城区道路等复杂环境下的车辆检测。更多还原

【Abstract】 The detection of the forward steering vehicles is one of the key tasks of the information perception process in Intelligent Transportation System. In this thesis, we investigate into the vehicle detection under the complicated traffic scenarios of urban roads and present a monocular-based method applying to vehicle detection.In the vehicle detection section, we first adopt a three-rank pyramidal Gauss filter to smooth the original image and use a Sobel 3*3 operator to gain the vertical edges of vehicle contour as global feature. After the imaging width threshold of vehicles is calculated, a so-called“voting”mechanism is applied to map the symmetrical axes of edges into a 1-D vector. After the study of experiment data and relevant academic technologies, we detect the vehicle symmetry axes based on Wavelet Transform Module Maximum(WTMM) theory. In order to generate the Region of Interests(ROI) for latter vehicle validation, we combine multi-features of vehicle as judgment bases to detect its external rectangle. Horizontal contour symmetry, bottom shadow and shape scale are utilized respectively to detect the left, right, bottom and upper boundary. Then, vehicle validation process is accomplished by local entropy and local gray scale symmetry estimate. Finally, The validated ROI area is filtered to obtain the ultimate detection results.The algorithm was tested under different traffic scenarios (e.g., simply structured highway, complex urban street, varying sunlight conditions), especially under the complicated traffic circumstance of urban roads. The approving experimental results show that the method is effective, stabilized and robust. Additionally, it is fast enough to be used in real-time Intelligent Transportation System. Thus, the method could be well used in vehicle detection under urban traffic scenarios.更多还原