基于智能手机的车辆检测与车距测量

发布时间：2018-03-02 12:36

本文选题：辅助驾驶系统　切入点：车辆检测　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,随着汽车保有量的增加,交通事故成为一个不得不重视的话题。汽车辅助驾驶系统(ADAS),特别是车辆防碰撞预警系统和车道保持辅助系统也变得越来越重要。其中前车检测和车距测量是车辆防碰撞预警系统的关键模块。在前车检测系统中,常见的传感器有高频雷达(毫米波)、超声波、红外激光雷达,摄像头等,每种传感器适应的场景不同,想要适应多种场景,汽车厂商和辅助驾驶系统提供商一般会融合多种传感器收集到的信息进行决策。由于基于视觉的前车检测系统成本低,信息丰富,且近年来计算机视觉领域的高速发展为其提供了技术基础,所以受到了越来越多相关领域研究人员的重视。基于计算机视觉的车辆检测系统主要基于车辆的表面特征,比如对称性,颜色,纹理,阴影,几何特征,车灯等,之后通过模版匹配或者机器学习方法对图像进行检测。基于视觉的车距测量方法主要有双目视觉测距和单目视觉测距两种,鉴于双目视觉测距系统成本较高,计算复杂,目前单目视觉测距仍然是主流。当前的辅助驾驶系统生产厂商会定制专门的硬件设备来实现车辆检测和车距测量,平台搭建费用很高,普及程度较低。本文的车辆检测和车距测量系统基于智能手机平台,包括Android和IOS两大主流平台,使用手机的摄像头录制的视频进行实时检测,大大降低了汽车辅助驾驶系统的使用成本,并且改进了车辆检测方法,使其能够适应多种复杂环境,包括阴雨天、城市道路、光线突变等,并且有效降低了系统的误检率,提高了准确度,使其有望得到普及。为了兼顾实时性和准确度,本文系统采用两步前车检测技术,即寻找车辆假设区域和验证车辆假设区域。具体地,通过基于车道先验的两步阈值法提取车底阴影,通过筛选规则对阴影区域进行筛选从而得到车辆假设区域,之后利用基于haar-like特征的Adaboost方法来对车辆假设区域进行验证,最后通过视频流特征对误检进行剔除并对车辆进行跟踪。本文还结合摄像头透视几何关系,提出了基于车道消失点的单目视觉测距方法,使得只需用户提供摄像头高度信息即可达到对前车距离进行准确估算的目的,可以作为有效的危险预警手段。实时性也是辅助驾驶系统的重要需求特性,该系统在保证准确率的同时,综合考虑了智能手机的特性,对该系统进行了优化,保证了系统运行的实时性。实验结果表明该系统具有较高的准确度和较低的误检率,并且能够在智能手机上实时运行,在晴朗、阴雨以及城市道路、高速道路等多种环境下表现良好,能够有效地应用到车辆检测和车距测量等汽车辅助驾驶系统中。
[Abstract]:In recent years, with the increase in car ownership, traffic accident has become an important topic. To automobile driver assistance system (ADAS), especially the vehicle anti collision warning system and lane keeping assist system has become more and more important. The front vehicle detection and distance measurement is the key module of vehicle anti collision warning system. In the front vehicle detection system, the common sensor has a high frequency radar (millimeter wave), ultrasonic, infrared laser radar, cameras, each kind of sensor to adapt to the different scenes, to adapt to a variety of scenes, car manufacturers and auxiliary driving system providers generally will integrate a variety of sensors to collect information for decision-making. Based on the vehicle in front the visual detection system of low cost, abundant information, and in recent years the rapid development in the field of computer vision which provides a technical basis, so has received more and more research in related fields Personnel attention. Computer vision based vehicle detection system is mainly based on the surface characteristics of the vehicle, such as symmetry, color, texture, shadow, geometric feature, lights, followed by template matching of image or machine learning method based on visual detection. The vehicle distance measurement includes binocular vision ranging and monocular vision two, given the cost of binocular vision ranging system of high computational complexity, the monocular vision is still the mainstream manufacturers. The auxiliary drive system will specifically customized hardware to realize vehicle detection and distance measurement platform, high cost, popularity is low. The distance of the vehicle detection and vehicle the measurement system based on intelligent mobile phone platform, including Android and IOS two mainstream platform, using a mobile phone camera to record the video in real-time detection, greatly reducing vehicle auxiliary driving The use of cost driving system, and improved the vehicle detection method, which can adapt to the complex environment, including the rainy days, city road, light changes, and effectively reduce the system error rate, improve the accuracy, which is expected to be popular. For both real time and accuracy, this paper uses the system the two step preceding vehicle detection technology, which is looking for vehicles that regional and regional specific assumptions. Verify that the vehicle, by extracting the shadows two step Lane threshold method based on prior to the shadow region so as to obtain the regional vehicle hypothesis screening by screening rules, after using the Adaboost method based on Haar-like feature to verify the hypothesis of regional vehicle finally, through the video features to remove the false detection and tracking of the vehicle. This paper also combines the X-ray camera geometry, proposed a single lane based on visual vanishing points Sleep ranging method makes users only need to provide the camera height information can achieve accurate estimation of the objective from the front of the car, can be used as the early warning method. Real time is an important demand characteristic of auxiliary driving system, the system accuracy in ensuring at the same time, considering the characteristics of Intelligent Mobile phone, the system optimized to ensure real-time operation of system. The experimental results show that the system has high accuracy and low false alarm rate, and can in the intelligent mobile phone real-time operation, in the sunny, rainy and city roads, high-speed road under a variety of environmental performance is good, can be effectively applied to vehicle detection and vehicle distance measurement in driver assistant system.

【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495;U463.6;TP391.41

【参考文献】