基于DM642车辆视频检测器

发布时间：2019-05-20 20:18

【摘要】：车辆是城市道路交通最重要的组成部分,因此开展车辆视频检测器的研究对城市智能交通系统的发展具有重要的意义。随着智能交通系统对产品小型化、低功耗、低成本且稳定可靠的需求日益强烈,开展嵌入式车辆视频检测器的研究更具有了重要的现实意义和应用价值。本论文基于TI公司的数字信号处理器(DSP)TMS320DM642平台,对嵌入式车辆视频检测算法进行了研究,设计与实现了嵌入式的视频车辆检测器。论文的主要内容和研究成果包括： 1.嵌入式算法设计：本文在虚拟线圈算法框架下,引入速度、方差和均值等计算简单的统计量,提出了基于虚拟线圈统计量的快速视频车辆检测算法。该算法首先基于块匹配原理计算虚拟线圈速度统计量,然后采用高阶统计量分块法来进行背景建模并得到前景图像的均值和方差统计量,最后通过统计量的逻辑运算来判断车辆到达、经过、离开、驻停等状态。该算法在满足实时性要求的前提下,对于光照、阴影等外界条件的干扰有较强的鲁棒性,适合在嵌入式平台上运行。实验结果表明,该算法在PC机上可以达到50帧/秒的检测速度,同时在理想条件下白天的检测率可以达到98%以上,夜间的检测率达到93%,在白天车辆密集且有阴影时检测率达到94%,满足实际使用需求。 2.程序实现与优化：本文以基于虚拟线圈统计量的车辆检测算法为软件核心,在DM642平台上实现了嵌入式视频车辆检测器。通过使用内联函数、展开循环、重排软件流水、利用乒乓机制传输数据等多种手段对程序进行了优化。实际道路的测试结果表明,本文研制的嵌入式车辆检测器能够达到20帧/秒的检测速度,达到了实时检测,同时车辆检测率白天可以达到96%,夜晚达到90%,误检率白天不高于2%,漏检率白天为4%,误检率夜晚不高于4%,漏检率夜晚为10%。测试结果表明,本文研制的基于DM642的视频车辆检测器能够对车辆进行实时的检测,同时也保证了检测的准确性,在检测速度和检测率两者之间的平衡性上表现出色,为以后嵌入式视频车辆检测器的研制与应用奠定了一定的基础。
[Abstract]:Vehicle is the most important part of urban road traffic, so the research of vehicle video detector is of great significance to the development of urban intelligent transportation system. With the increasing demand of intelligent transportation system for miniaturization, low power consumption, low cost, stability and reliability, the research of embedded vehicle video detector has important practical significance and application value. In this paper, based on the digital signal processor (DSP) TMS320DM642 platform of TI Company, the embedded vehicle video detection algorithm is studied, and the embedded video vehicle detector is designed and implemented. The main contents and research results of the paper include: 1. Embedded algorithm design: in the framework of virtual coil algorithm, this paper introduces simple statistics such as speed, variance and mean, and proposes a fast video vehicle detection algorithm based on virtual coil statistics. Firstly, the virtual coil velocity statistics are calculated based on the block matching principle, and then the background modeling is carried out by using the high-order statistics block method, and the mean and variance statistics of the foreground image are obtained. Finally, the logical operation of statistics is used to judge the state of vehicle arrival, passing, leaving, stopping and so on. On the premise of meeting the real-time requirements, the algorithm has strong robustness to the interference of light, shadow and other external conditions, and is suitable for running on embedded platforms. The experimental results show that the detection speed of the algorithm can reach 50 frames per second on PC, and the detection rate can reach 98% in daytime and 93% in night under ideal conditions. When the vehicles are dense and shaded during the day, the detection rate reaches 94%, which meets the actual demand. 2. Program implementation and optimization: this paper takes the vehicle detection algorithm based on virtual coil statistics as the software core, and implements the embedded video vehicle detector on DM642 platform. The program is optimized by using inline function, expanding loop, rearranging software pipeline, using ping-pong mechanism to transmit data and so on. The test results of the actual road show that the embedded vehicle detector developed in this paper can achieve the detection speed of 20 frames per second and achieve real-time detection. At the same time, the vehicle detection rate can reach 96% during the day and 90% at night. The false detection rate is not more than 2% during the day, the missed detection rate is 4% during the day, the false detection rate is not more than 4% at night, and the missed detection rate is 10% at night. The test results show that the video vehicle detector based on DM642 can detect the vehicle in real time, at the same time, it also ensures the accuracy of the detection, and performs well in the balance between the detection speed and the detection rate. It lays a certain foundation for the development and application of embedded video vehicle detector in the future.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：U495;TP368.1

【参考文献】