基于双目视觉的无人车行驶障碍物定位跟踪方法研究

发布时间：2018-04-25 04:31

本文选题：无人车 + 双目视觉　；参考：《长安大学》2016年硕士论文

【摘要】：无人车是一种能够通过传感器感应识别周围环境,主动地进行避障和行驶路径规划的智能汽车。无人车在减少交通事故和交通违规行为,减轻驾驶员劳动强度,以及合理规划道路资源的分配使用方面具有很大优势。无人车对周围环境和障碍物的主动感知与监测技术的研究是发挥无人车在交通系统中优势的关键点。基于双目立体视觉的无人车环境识别技术作为无人车对周围环境障碍物识别的重要方法已成为研究的重点与热点。因此,论文基于双目立体视觉测量技术对无人车行驶中障碍物的定位与跟踪方法进行了研究,主要研究内容有:(1)建立可应用于无人车的双目立体视觉障碍物位置测量模型。通过建立包含畸变系数的摄像机针孔模型和由两个摄像机组成的双目摄像机系统模型,并结合无人车坐标系统、双目摄像机在无人车上的安装关系研究提出适用于无人车障碍物测量的双目立体视觉障碍物位置测量模型。(2)搭建并标定双目立体视觉障碍物跟踪测试平台。使用USB接口摄像机和安装架搭建双目立体测量试验平台并通过经典张正友平面标定法获得摄像机内参矩阵和外参矩阵。然后对双目立体测量试验平台进行立体标定得到左右摄像机位置的旋转、平移矩阵并根据获得的参数和测量模型求解得出障碍物的三维深度信息矩阵。(3)提出基于深度信息与直方图反向投影法相结合的CamShift障碍物跟踪算法。针对传统CamShift跟踪算法需要手动选择初始搜索位置且障碍物与背景色调相近时容易跟丢的缺点,研究并实现基于深度信息的改进CamShift障碍物跟踪算法。(4)测量双目立体视觉测试平台的精度并对论文提出的改进跟踪算法进行试验研究和验证。通过测量给定障碍物的距离试验和实时跟踪特定场景运动障碍物的试验获得双目立体视觉测试平台的测量精度并验证论文提出的基于深度信息的改进CamShift障碍物跟踪算法的可行性。论文以建立适用于无人车的双目立体测量模型、改进传统跟踪算法并进行试验验证为研究思路,研究提出有效的无人车障碍物定位计算模型和基于深度信息的跟踪算法,具有重要的理论和实用价值。
[Abstract]:Unmanned vehicle (UAV) is a kind of intelligent vehicle which can recognize the surrounding environment and actively avoid obstacles and plan the driving path by sensors. Unmanned vehicles have great advantages in reducing traffic accidents and traffic violations, reducing drivers' labor intensity, and rationally planning the allocation and use of road resources. The research on the active perception and monitoring of the surrounding environment and obstacles is the key point to give full play to the advantages of the unmanned vehicle in the traffic system. Binocular stereo vision based environment recognition technology for unmanned vehicles as an important method to identify obstacles around the environment has become the focus and focus of research. Therefore, based on binocular stereo vision measurement technology, this paper studies the location and tracking method of obstacles in driverless vehicles. The main content of this paper is to establish a binocular stereo vision obstacle position measurement model which can be applied to unmanned vehicles. By establishing the camera pinhole model with distortion coefficient and the binocular camera system model composed of two cameras, and combining with the unmanned vehicle coordinate system, Study on the installation relationship of binocular camera in unmanned vehicle A binocular stereo vision obstacle position measurement model. 2) A binocular stereo vision obstacle tracking and testing platform is established and calibrated. The binocular stereo measurement test platform was built by using USB interface camera and mounting frame, and the camera inner and outer parameter matrices were obtained by the classical calibration method of Zhang Zhengyou plane. Then the binocular stereo measurement test platform is calibrated to get the rotation of the left and right camera position. Based on the translation matrix and the obtained parameters and measurement model, the 3D depth information matrix of the obstacle is obtained. (3) A CamShift obstacle tracking algorithm based on the combination of depth information and histogram reverse projection method is proposed. The traditional CamShift tracking algorithm needs to manually select the initial search position and the obstacles are easily lost when the obstacles are close to the background hue. An improved CamShift obstacle tracking algorithm based on depth information is studied and implemented to measure the accuracy of binocular stereo vision test platform. The improved tracking algorithm proposed in this paper is tested and verified. The measurement accuracy of the binocular stereo vision test platform is obtained by measuring the distance test of a given obstacle and the experiment of tracking the moving obstacle in a specific scene in real time. The improved CamShift obstacle and the improved CamShift obstacle based on depth information proposed in this paper are verified. The feasibility of tracing algorithm. Based on the idea of establishing a binocular stereo measurement model suitable for unmanned vehicle, improving the traditional tracking algorithm and verifying it through experiments, this paper presents an effective obstacle location calculation model and a tracking algorithm based on depth information. It has important theoretical and practical value.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U463.6

【参考文献】