基于OpenCV运动目标检测与跟踪方法研究

发布时间：2017-12-31 01:26

本文关键词：基于OpenCV运动目标检测与跟踪方法研究　出处：《沈阳航空航天大学》2017年硕士论文　论文类型：学位论文

【摘要】：视频中运动目标检测和跟踪是计算机视觉和模式识别领域的研究热点,在智能视频监控系统、人工智能、视觉导航等方面有着广泛的应用。本文以实际应用为背景,专注于目标检测与跟踪方法的研究。针对在固定背景(摄像头固定)下检测出运动的目标并实时进行标记跟踪。本文在对传统的跟踪算法深入研究的基础上,发现传统的方法存在着一些问题,不能满足在不同环境下对于检测跟踪的速度和精度上的要求。尤其是在复杂环境和多目标环境下,跟踪效果较差。因此,本文做了以下的工作:(1)在运动目标的检测部分,传统的检测算法主要有帧间差分法、背景减除法和光流法。在充分分析对比了各种算法的优缺点后,发现传统的检测算法在不同的环境下都不能实现很好的检测效果。因此,本文在基于传统的运动目标检测算法的基础上,实现了将背景减除法和背景建模相结合的目标检测算法进行运动目标的检测。在不同的环境下,实现了对运动目标比较理想的检测效果。(2)在运动目标的跟踪部分,如今使用最多的目标跟踪算法有MeanShift算法、粒子滤波算法以及Kalman算法。由于背景干扰、混乱、遮挡以及目标快速移动,传统的跟踪算法存在着跟踪漂移现象。因此,本文在深入研究粒子滤波跟踪算法的基础上,通过加入目标的空间位置分布信息,进行了改进,提出了基于空间位置--颜色直方图的粒子滤波跟踪算法。最后,通过大量的实验对比分析了不同算法的处理效果,验证了改进后算法的有效性和鲁棒性。对于现代智能化视频监控系统的发展及应用有着重要的意义和实用价值。
[Abstract]:Video moving target detection and tracking is a hot research field of computer vision and pattern recognition in the intelligent video surveillance system, artificial intelligence, has been widely used in visual navigation and so on. Based on the practical application and research focus on the methods of detecting and tracking targets. Aiming at the fixed background (camera) detection the moving target and the real-time marking tracking. Based on the traditional tracking algorithm on the basis of the in-depth study found that the traditional method has some problems, can not meet the environment in different speed and accuracy for the detection of the tracking requirements. Especially in the complex environment and multi target environment, tracking effect is poor therefore, this paper has done the following work: (1) in the detection of the moving target, the traditional detection algorithms are mainly inter frame difference method, background subtraction and optical flow method. In the full analysis Comparing the advantages and disadvantages of various algorithms, found that the traditional detection algorithms in different environments can achieve good detection effect. Therefore, based on the traditional moving object detection algorithms on the detection method of background subtraction and background modeling combined target detection algorithm for moving target in a different environment, to achieve a better detection effect of the moving target. (2) in the moving target tracking part, now most of the target tracking algorithm with MeanShift algorithm, particle filter algorithm and Kalman algorithm. The background interference, chaos, fast moving target tracking and occlusion, the traditional algorithms exist the tracking drift phenomenon. Therefore, based on the in-depth study of the particle filter tracking algorithm, the spatial distribution of information to the target, has been improved, based on space Particle filter tracking algorithm -- color histogram. Finally, through experiments and analysis the effect of different algorithms, which verifies the effectiveness and robustness of the improved algorithm. The development and application of modern intelligent video monitoring system has important significance and practical value.

【学位授予单位】：沈阳航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】