基于视频的交通事故自动判别算法研究

发布时间：2018-11-02 09:16

【摘要】：进入二十一世纪以来，我国社会和经济得到快速发展，城市化进程不断加快，城市道路的交通压力也在不断增大，城市交通拥堵现象愈发严重，且机动车数量的日益增加导致城市道路交通事故频发。传统的交通事故检测和查阅通常是通过人工监测的方式进行的，但这种方法效率低、实时性较差。随着智能交通的迅速发展以及计算机视觉技术的广泛应用，利用视频图像处理技术对道路监控视频进行实时分析、智能检测交通事故、获取交通信息成为了研究的热点。对交通事故的实时检测不仅可以减少警力资源的浪费，而且对于提高交通事故处理效率有着重大的意义。本文主要基于视频对交通事故的自动判别算法进行研究，论文的主要研究内容如下：首先介绍了运动目标检测的技术难点，归纳了初始化背景模型的常用方法，并利用均值法背景建模方法提取背景图像，获得初始背景之后，通过背景更新算法使背景及时更新到当前状态；然后利用背景差分法提取差分图像，并对车辆阴影进行去除后进行连通区域标定，实现运动目标的检测。通过进行实验并进行效果分析，验证了算法的有效性。运动目标跟踪算法部分，首先介绍了卡尔曼滤波器的工作原理，应用卡尔曼滤波对运动目标进行跟踪与匹配，利用运动前景的质心距离和面积大小作为匹配参数；运动目标特征提取部分，首先对摄像机标定算法进行分析，然后根据前一章运动目标检测和匹配跟踪的结果，通过提取前景目标的运动信息，计算目标的速度、行驶方向和轨迹等特征。通过卡尔曼滤波预测运动目标下一帧质心点坐标，并与检测到的前景质心位置进行比较，，判断检测到的前景是否为重合的前景，如果两个前景目标发生重合，则发生了交通冲突。然后，对交通冲突作进一步判断，提出了综合车辆减速度、行驶方向变化率和时间参数的自动判别交通事故的算法，对车辆造成的遮挡情况和伪碰撞现象进行识别，最终判断交通事故是否发生。最后通过实验进行验证，证明了判别算法的有效性，同时分析了实验存在的误差。
[Abstract]:Since the 21 century, the society and economy of our country have been developed rapidly, the process of urbanization has been quickened, the traffic pressure of urban roads is also increasing, and the phenomenon of urban traffic congestion is becoming more and more serious. And the increasing number of motor vehicles leads to frequent urban road traffic accidents. The traditional method of traffic accident detection and inspection is usually carried out by manual monitoring, but this method is inefficient and poor in real time. With the rapid development of intelligent transportation and the wide application of computer vision technology, real-time analysis of road surveillance video using video image processing technology, intelligent detection of traffic accidents, access to traffic information has become a research hotspot. The real-time detection of traffic accidents can not only reduce the waste of police resources, but also improve the efficiency of traffic accidents. The main contents of this paper are as follows: firstly, the technical difficulties of moving target detection are introduced, and the common methods of initializing background model are summarized. The background image is extracted by the mean method, and the background is updated to the current state by the background updating algorithm. Then the background difference method is used to extract the differential image, and then the connected region is calibrated after removing the shadow of the vehicle, and the moving object detection is realized. The effectiveness of the algorithm is verified by experiment and effect analysis. In the part of moving target tracking algorithm, firstly, the working principle of Kalman filter is introduced. Kalman filter is used to track and match moving target, and the centroid distance and area of moving foreground are used as matching parameters. In the part of feature extraction of moving targets, the camera calibration algorithm is analyzed first, and then according to the results of moving target detection and matching and tracking in the previous chapter, the velocity of the target is calculated by extracting the moving information of the foreground target. Driving direction and trajectory and other characteristics. The coordinates of the next frame centroid of moving target are predicted by Kalman filter, and compared with the detected centroid position of foreground, to judge whether the detected foreground is the same prospect, if the two foreground targets overlap, Traffic conflicts occur. Then, the traffic conflict is judged further, and an algorithm is proposed to automatically distinguish traffic accidents by synthesizing vehicle deceleration, direction change rate and time parameters, and to identify the occlusion and pseudo-collision caused by vehicles. Finally determine whether the traffic accident occurred. Finally, the validity of the discriminant algorithm is verified by experiments, and the error of the experiment is analyzed.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U491.31;TP391.41

【参考文献】