基于视频的运动目标检测与跟踪算法的研究

发布时间：2018-09-17 09:44

【摘要】：近年来随着科学水平的发展与计算机技术的进步,运动目标检测与跟踪技术取得了许多成果,期间涌现出不少优秀算法,实际应用领域也愈加宽广。然而,在复杂环境下,依旧存在着很多问题。针对运动目标检测与跟踪算法存在的不足,本文提出了一种新的运动目标检测与跟踪算法。总的来说,本文主要研究工作和创新性如下:1、图像处理基础知识的介绍。本文概括介绍了图像平滑处理、边缘检测和数学形态学处理的相关知识,针对本文所涉及的一些处理技术,包括中值滤波、Canny算子、膨胀腐蚀等进行了详细介绍。2、运动目标检测算法的研究。首先,介绍了传统的光流法、帧间差分法及背景差分法的基础概念及公式原理,并对其中几种代表性的算法进行分析对比,列出了它们的优缺点。然后,在此基础上提出了一种基于视觉背景提取模型的新算法——IDVibe算法。该算法分别从模型建立、模型匹配、模型更新及前景分割四个方面进行改进,通过融入三帧差分法的思想,有效地解决了“鬼影”、光照变化等问题。最后,通过实验仿真可以得到,本文提出的检测算法能更好地适应动态复杂的环境,有良好的检测效果和鲁棒性。3、运动目标跟踪算法的研究。本文以卡尔曼滤波跟踪算法、Mean shift算法和粒子滤波算法为基础,提出了一种融合多特征与Mean Shift的粒子滤波跟踪算法。首先,通过IDVibe算法对运动目标进行检测、定位。然后,融合颜色、纹理及边缘的特征信息进行模型匹配,实现粒子滤波跟踪。最后,运用Mean shift算法的收敛性,将粒子重新聚集到真实目标附近,实现运动目标的跟踪。通过实验结果可以证明,本文提出的跟踪算法能达到较好的跟踪效果,对比以往单一的跟踪算法有更高的准确性与实时性。
[Abstract]:In recent years, with the development of science and computer technology, many achievements have been made in moving target detection and tracking technology. During this period, many excellent algorithms have emerged, and the practical application field has become wider and wider. However, in the complex environment, there are still many problems. A new algorithm for moving target detection and tracking is proposed in this paper. In general, the main work and innovation of this paper are as follows: introduction of basic knowledge of image processing. This paper summarizes the knowledge of image smoothing, edge detection and mathematical morphology processing. Some processing techniques, including median filter and Canny operator, are discussed in this paper. The paper introduces in detail the. 2. 2, the research of moving target detection algorithm. Firstly, the basic concepts and formula principles of traditional optical flow method, inter-frame difference method and background difference method are introduced, and several representative algorithms are analyzed and compared, and their advantages and disadvantages are listed. Then, a new algorithm based on visual background extraction model, IDVibe algorithm, is proposed. The algorithm is improved from four aspects: model establishment, model matching, model updating and foreground segmentation. By incorporating the idea of three-frame difference method, the problems of "ghost" and illumination change are effectively solved. Finally, through the experiment simulation, we can get that the detection algorithm proposed in this paper can better adapt to the dynamic and complex environment, have good detection effect and robustness. 3, the research of moving target tracking algorithm. Based on the Kalman filter tracking algorithm, mean shift algorithm and particle filter algorithm, a particle filter tracking algorithm combining multiple features and Mean Shift is proposed in this paper. First, the moving targets are detected and located by IDVibe algorithm. Then, the feature information of color, texture and edge are fused to match the model to realize particle filter tracking. Finally, using the convergence of the Mean shift algorithm, the particles are reassembled near the real target to achieve the tracking of moving targets. The experimental results show that the proposed tracking algorithm can achieve better tracking effect and has higher accuracy and real-time performance than the previous single tracking algorithm.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】