基于视觉的汽车前方碰撞预警系统研究

发布时间：2018-01-12 14:03

本文关键词：基于视觉的汽车前方碰撞预警系统研究　出处：《湖南大学》2016年硕士论文　论文类型：学位论文

【摘要】：基于视觉的汽车前方碰撞预警(V-FCW)系统是汽车安全辅助驾驶技术领域最为典型的应用之一。汽车前方碰撞预警(FCW)系统主要利用传感器获取信息并处理,准确判断前方可能发生的碰撞危险情况并对驾驶人给与警告,与目前多采用雷达传感器的汽车前方碰撞预警系统相关研究不同,基于视觉的汽车前方碰撞预警系统使用摄像头作为获取信息的环境感知工具。视觉传感器具有模式识别及车道线检测等其他传感器不具备的优势,但同时也面临着障碍物识别及测距等方面的巨大技术挑战。基于此研究背景,本文设计了一套V-FCW系统新的车辆检测、车辆测距方法和安全距离计算算法,并在通过仿真分析验证其效果。在对汽车前方碰撞预警安全距离模型分析的基础上,本文根据车辆的运行状态和制动规律,充分考虑了影响车辆制动的驾驶人因素和环境因素,建立了新的汽车安全距离模型,同时对该模型中的参数进行了说明和分析。结合其他模型算法利用MATLAB软件进行仿真分析,结合不同状态下最小安全车距与速度和附着系数的变化规律。验证了该模型能够更好的地反映真实车辆运行状况,降低了计算安全车距存在的较大偏差。本文在车辆检测识别方面作出改进。利用类Haar和改进的Ada Boost算法的对前方车辆进行检测,首先利用基于积分图的思想计算图像的扩展类Haar特征;然后使用提取的类Haar特征值向量集改进Ada Boost算法,文中提出了改进措施:利用SVM代替初步分类器Ada Boost算法(SVM-Ada Boost)增强了分类器的分类能力。通过实验验证文中所提出的方法在训练所需时间和识别性能方面取得了比传统方法更好的效果。本文在车辆测距方面,使用单目摄像机利用小孔成像的测距原理,通过对摄像机内部参数的标定,根据建立的测距模型实现前后车辆之间的距离测量,并使用行车记录仪的视频对前方车辆距离进行测量实验。
[Abstract]:Vision-based vehicle front collision warning system is one of the most typical applications in the field of vehicle safety driving assisted driving technology. The system mainly uses sensors to obtain information and process. It is different from the current research on the vehicle front collision warning system that radar sensors are used to accurately judge the possible collision hazard and give warning to the driver. The vision-based vehicle front collision warning system uses the camera as an environment sensing tool to obtain information. The visual sensor has the advantages of pattern recognition and lane detection and other sensors do not have the advantages. But at the same time, it also faces huge technical challenges in obstacle recognition and ranging. Based on this research background, this paper designs a new set of V-FCW system vehicle detection. Vehicle ranging method and safety distance calculation algorithm, and through simulation analysis to verify its effect. On the basis of the analysis of the vehicle front collision warning safety distance model, this paper according to the vehicle running state and braking law. Considering the factors of driver and environment, a new vehicle safety distance model is established. At the same time, the parameters of the model are explained and analyzed. Combined with other model algorithms, MATLAB software is used to simulate the model. Combined with the minimum safety vehicle distance, speed and adhesion coefficient under different conditions, it is verified that the model can better reflect the actual vehicle operation status. The large deviation of calculating the safe vehicle distance is reduced. In this paper, the vehicle detection and identification are improved. The forward vehicle is detected by using the similar Haar and the improved Ada Boost algorithm. Firstly, the extended Haar feature of the image is calculated based on the idea of integral graph. Then the Ada Boost algorithm is improved by using the extracted Haar like eigenvalue vector set. The improvement measures are put forward in this paper: using SVM instead of the initial classifier Ada Boost algorithm, SVM-Ada boost (SVM-Ada boost). The experimental results show that the proposed method is more effective than the traditional method in training time and recognition performance. Based on the calibration of camera internal parameters, the distance between front and rear vehicles can be measured by using the ranging principle of monocular camera by using the principle of pinhole imaging. The video of the CarLog is used to measure the distance of the vehicle in front of the vehicle.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U463.6

【相似文献】