【作者】 王鹏飞；

【导师】 王晋疆；

【作者基本信息】 天津大学 ， 光学工程， 2018， 硕士

【摘要】 针对全景环视图存在拼接线段断裂的问题,本文在阅读大量国内外文献的基础上搭建了全景环视系统,分析了不同拼接区域及不同融合算法对生成全景环视图的影响,在此基础上研究了“U形”车位的检测与识别算法,为基于视觉传感器的车位检测技术提供了参考。本文工作主要分为以下几个方面:(1)介绍了鱼眼摄像机模型以及鱼眼摄像机的内外参标定原理,其中,鱼眼摄像机模型近似于等距投影模型,推导了空间中一点成像到鱼眼图像上一点的过程,采用张正友标定算法标定相机内参,计算四路环视相机图像坐标系与同一世界坐标系之间的转换关系进行相机外参的标定。(2)分析了三角函数法与棋盘格标定法的俯视变换效果,对四路环视相机进行俯视变换,获取相机俯视图。为了加快拼接速度,采用基于像素查找表的拼接方法,将全景环视图像与俯视图之间的转换关系保存,每次拼接时只需要将图像坐标系中的像素点搬运到俯视图像对应像素位置处。研究不同拼接区域选择对拼接缝的影响,最终采用加权平均融合算法对拼接区域进行融合。(3)根据车位检测算法的要求,分析了Hough变换与基于LSD(Line Segment Detector)的直线检测算法对车位检测识别精度的影响,针对停车场景中存在大量的“U形”车位,设计算法过滤干扰线、提取停车位角点、拟合有效车位。由于车辆在行驶过程中,经常会出现漏检现象,采用光流跟踪算法对车位进行跟踪,确定了车位检测系统的设计流程。(4)根据车位检测的要求搭建基于全景环视的车位检测系统,为了验证本文算法的有效性,对不同泊车场景进行实验测试,主要包括晴天与雨天、地下车位与地上车位,对检测识别效果进行实验测试与对比分析。实验结果表明,本文车位检测算法适用于大多数天气情况以及泊车场景,能够有效检测到车位并减少漏检和误检。更多还原

【Abstract】 Aiming at the shortcomings of the broken stitching line segments in the panoramic image,based on the extensive understanding of the domestic and foreign literatures,the Around View Monitor(AVM)systems is built.The effects of different fusion algorithms and different splicing areas on the generation of AVM image are analyzed.The detection and recognition algorithm of "U-shaped" parking slot based on AVM image is studied,which provides a reference for parking slot detection technology based on vision sensor.The work is composed of the following aspects:(1)The fisheye camera model and the calibration principle of internal and external parameters of the fisheye camera is introduced.Among them,the fisheye camera model approximates the equidistant projection model,and deduces the process of imaging a point in the space to the fisheye image.The Zhang Zhengyou calibration algorithm is used to calibrate the camera internal parameters,the conversion relationship between the four around-view camera image coordinate systems and the same world coordinate system is calculated to calibrate the camera external parameters.(2)The effect of the top-view transformation of the trigonometric function method and the checkerboard calibration method is analyzed.The four around-view camera is transformed into a top view to obtain a top view of the camera.In order to speed up the splicing speed,the splicing method based on the pixel lookup table is used to save the conversion relationship between the around-view image and the top view image,and only need to carry the pixel points in the image coordinate system to the top-view image corresponding pixel position for each splicing.The influence of different splicing area selection on the splicing seam was studied.Finally,the splicing area was fused by weighted average fusion algorithm.(3)According to the requirements of the parking slot detection algorithm,the influence of Hough transform and LSD(Line Segment Detector)based line detection algorithm on the detection accuracy of parking slot detection is analyzed.For the large number of "U-shaped" parking slots in the parking lot,the algorithm is designed to filter out the interference line and extract corners of the parking slot,fits the effective parking slot.When the vehicle is in the process of running,the phenomenon of missed detection often occurs.The parking slot is tracked by optical flow tracking algorithm,and the design flow of the parking slot detection system is determined.(4)According to the requirements of parking slot detection,an parking slot detection system based on panoramic around view is built.In order to verify the effectiveness of the algorithm,experimental tests are carried out on different parking scenarios,including sunny and rainy days,underground parking slot and above-ground parking slot.Experimental testing and comparative analysis were carried out on the detection and recognition effects.The experimental results show that the parking slot detection algorithm is suitable for most weather and parking scenarios,and can effectively detect the parking slot and reduce missed detection and false detection.更多还原