基于无人机视频的运动车辆检测研究

发布时间：2018-01-08 17:12

本文关键词：基于无人机视频的运动车辆检测研究　出处：《北京交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着无人机研发技术日益成熟,无人机在民用领域各行业的应用开发逐渐成为热门的研究课题。无人机具有低成本、易部署、高机动性、大视角范围和统一尺度等优势,可广泛应用于交通监控和交通信息采集。本文综合考虑无人机在交通视频采集过程中具有悬停、巡航不同的飞行状态,针对无人机视频图像的特性,设计开发了基于无人机视频的运动车辆检测算法,并与帧差法、混合高斯算法、光流法等多种运动目标检测算法进行对比研究,最后通过实验证明了本文算法检测精度较高,并能满足实时应用的需要。本文的主要研究内容如下:(1)本文分析了无人机的发展前景和行业应用的开发优势,总结了无人机在交通领域的应用研究现状,详细归纳了相关运动平台视频目标检测技术和图像处理技术,并对基于视频的运动目标检测的主流算法进行了详细分析。(2)在考虑了无人机悬停和巡航不同飞行状态下视频处理需求的基础上,开发了基于无人机视频的运动车辆检测算法,实现过程分为:①设置感兴趣区域选择运动车辆检测的范围;②引入ViBe背景更新算法进行运动前景区域提取,改进了ViBe算法使用固定阈值和存在鬼影问题的缺陷,提高了算法对无人机视频运动目标检测的效果;③通过形态学优化处理和车辆目标识别完成了运动车辆的检测。(3)通过定性、定量的实验分析,将帧差法、混合高斯算法和光流法三种主流的运动目标检测算法与本文算法进行了比较,证明本文算法在无人机悬停和巡航状态下均具有较高检测精度(90.25%)和实时的计算速度(42.69fps),能够有效地进行基于无人机视频的运动车辆检测,实现实时应用。本研究对于丰富基于无人机视频的车辆检测和交通信息采集方法具有重要的参考意义,有助于无人机在交通领域应用的推广。
[Abstract]:With the development of UAV technology is increasingly mature, no application development in the field of civil UAV industry has gradually become a hot research topic. The UAV has the advantages of low cost, easy deployment, high mobility, large scope and unified scale and other advantages, can be widely used in traffic monitoring and traffic information collection. Considering the UAV with hovering in the traffic video capture process, different cruise flight, according to the characteristics of UAV video image, design and development of vehicle UAV video detection algorithm based on difference method, and with the frame, the mixed Gauss algorithm, comparative study of multiple moving targets detection algorithm of optical flow method, the experiment proved that the accuracy is higher the detection algorithm, and can meet the needs of real-time applications. The main contents of this paper are as follows: (1) this paper analyzes the development of UAV's development prospects and industry application No advantage, summarizes the application research situation in the field of transportation machine, sums up the relevant motion platform video object detection technology and image processing technology, and the mainstream of the moving target detection algorithm based on video analysis in detail. (2) in the UAV suspension based video processing and cruise stop needs different flight conditions on the development of the UAV video vehicle detection algorithm based on the realization process is divided into: setting ROI selection range of vehicle detection; the introduction of ViBe background updating algorithm of motion foreground domain extraction improved, and the existence of ghost defects using fixed threshold ViBe algorithm, improved algorithm the UAV video moving target detection; through the morphological processing and optimization of vehicle target recognition complete vehicle detection. (3) through qualitative and quantitative experiments Analysis of the frame difference method, the moving target detection algorithm and the algorithm of hybrid Gauss method and optical flow method three kinds were compared to prove that this algorithm has high detection accuracy in UAV hover and cruise state (90.25%) and real time computing speed (42.69fps), can effectively carry out vehicle UAV video detection based on real-time application. This research is to enrich the UAV video vehicle detection and traffic information collection method based on has the important reference significance, contribute to the promotion of human free applications in the transport sector.

【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;U495

【参考文献】