基于单目视频的车辆对象轮廓清晰化及速度测定方法研究

发布时间：2018-05-25 17:03

本文选题：视频质量评价 + 车辆轮廓清晰化　；参考：《昆明理工大学》2015年硕士论文

【摘要】：智能交通系统(Intelligent Transportation System)是未来交通系统的主流,先进的交通控制和计算机科学有效的结合,将高效服务于人类,极大提高监管以及违章肇事处理效率。目前,监管部门对车辆的监控执法,主要通过车牌自动识别系统识别车辆牌照,以此作为违法证据对其进行责任的追究。而对于牌照故意遮挡、牌照有牌不挂以及使用套牌等违章行为,现有车牌自动识别系统无法识别,需要通过人工辨别进行处罚,耗时费力。解决这一问题成为智能交通中亟待解决的热点和难点。本文的研究内容是分布式视频车辆同一性时空关联检索中的基础研究工作,为视频车辆同一性时空关联检索提供了重要的基础技术支持。本文的主要研究点有以下几个方面：1、视频样本建立及质量分析评价。针对高速公路卡口角度不同、道路环境不同、天气和光线的时常变化等因素,均会影响到车辆对象的提取以及进一步清晰化处理。故建立基于不同卡口的视频样本库,进行视频质量评价,采用基于ROI运动目标区域的视频质量评价方法,针对视频质量评价的结果进行对应的视频预处理工作,使视频质量评价有效关联视频的预处理。2、车辆对象的提取及轮廓清晰化。首先采用了Gamma光照补偿、高斯滤波、双边滤波、拉普拉斯锐化、直方图均衡化等预处理方法,然后对预处理后的视频进行混合高斯背景建模提取出目标车辆,用基于HSV颜色空间的阴影消除算法,消除阴影得到完整的车辆对象,最后利用序列帧间运动补偿的方法进行轮廓修复,并引入轮廓修复评价函数,控制修复的合理性和有效性,确保车辆对象轮廓清晰完整。3、基于车道线的视频测速。利用高速公路中车道线已知标准的信息,建立以车道线为基准的测速区域,通过车辆对测速区域进出边的撞边算法计算撞边帧数,根据视频帧率换算出时间,从而计算出车辆的行驶速度。
[Abstract]:Intelligent Transportation system is the mainstream of traffic system in the future. The combination of advanced traffic control and computer science will efficiently serve human beings and greatly improve the efficiency of supervision and handling of violations. At present, the monitoring and enforcement of vehicles by the regulatory authorities, mainly through the license plate automatic recognition system to identify the vehicle license plate, as evidence of violations of the law to investigate their responsibilities. However, the existing license plate automatic recognition system can not recognize the license plate intentionally, the license plate does not hang and the license plate is used, so it needs to be punished by manual discrimination, which is time-consuming and laborious. Solving this problem has become a hot and difficult problem in intelligent transportation. The research content of this paper is the basic research work in distributed video vehicle identity time and space link retrieval, which provides important basic technical support for video vehicle identity time and space link retrieval. The main research points of this paper are as follows: 1, video sample establishment and quality analysis and evaluation. In view of the different angle of highway bayonet, the different road environment, the frequent change of weather and light, etc., all these factors will affect the extraction of vehicle object and the further clear processing. Therefore, the video sample database based on different bayonets is established to evaluate the video quality, and the video quality evaluation method based on the moving target area of ROI is adopted to carry out the corresponding video preprocessing work according to the results of the video quality evaluation. It makes the video quality evaluation effective correlation video preprocessing. 2, vehicle object extraction and contour clarity. Firstly, Gamma illumination compensation, Gao Si filtering, bilateral filtering, Laplacian sharpening, histogram equalization and other preprocessing methods are used, and then the target vehicle is extracted by hybrid Gao Si background modeling. The shadow elimination algorithm based on HSV color space is used to eliminate the shadow to obtain the complete vehicle object. Finally, the contour restoration is carried out by using the method of motion compensation between sequence frames, and the contour repair evaluation function is introduced to control the rationality and effectiveness of the restoration. Ensure vehicle object profile is clear and complete. 3. Video speed measurement based on lane line. Based on the known standard information of freeway lane line, the speed measuring area based on lane line is established, and the number of frames of collision edge is calculated by the algorithm of vehicle collision edge in and out of the speed measuring area, and the time is converted according to video frame rate. Thus the speed of the vehicle is calculated.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495;TP391.41

【参考文献】