视频图像动目标检测与定位方法的研究

发布时间：2018-01-05 09:39

本文关键词：视频图像动目标检测与定位方法的研究　出处：《大连海事大学》2017年硕士论文　论文类型：学位论文

【摘要】：视频图像动目标检测一直是计算机视觉领域的一个重要发展方向,并且在海事闭路电视监控系统、智能交通等领域有广泛的应用。常用的检测算法有:光流法、统计中值法、均值法、帧间差分法等,但是由于慢速运动目标与背景相似,传统的目标检测算法对于慢速运动目标不敏感,容易出现漏检的情况。针对这一问题,本文分析了传统检测算法的特点,选择合适的图像预处理技术,并对传统的帧间差分法的不足进行改进。通过分析前后目标的相关度,然后利用二分收敛算法逼近最佳帧间隔。动目标检测的准确性也取决于背景的精确程度,本文提出了基于时空联合的背景迭代算法,通过自适应的加权因子和更新系数,不断对背景进行修正,然后利用差分法获得运动目标,最后利用快速局部阈值法对图像进行分割得到目标二值化图像。视觉定位算法由于其较高的应用价值和良好的发展前景,逐渐成为智能监控领域的一个研究热点。本文建立成像模型,推导相应的投影矩阵,并利用三点标定法获得摄相机参数,实现单目摄像头定位。实验证明本文的改进帧间差分法和基于时空联合的目标检测方法对于慢速目标同样具有良好的检测效果,并且基于时空联合的检测方法具有较好的鲁棒性,优越性更加突出。本文的视觉定位算法实现简单,误差较小,实验证明了该定位方法的有效性。
[Abstract]:Video image moving target detection has been an important development direction in the field of computer vision, and has been widely used in maritime closed-circuit television monitoring system, intelligent transportation and other fields. The commonly used detection algorithms are: optical flow method. Statistical median method, mean value method, inter-frame difference method, etc., but because the slow moving object is similar to the background, the traditional target detection algorithm is not sensitive to the slow moving target. It is easy to miss detection. In view of this problem, this paper analyzes the characteristics of the traditional detection algorithm, and selects the appropriate image preprocessing technology. The traditional inter-frame difference method is improved. By analyzing the correlation between the targets before and after, and then using dichotomous convergence algorithm to approach the best frame interval, the accuracy of moving target detection depends on the accuracy of the background. In this paper, a background iterative algorithm based on spatio-temporal union is proposed. The background is modified continuously by adaptive weighting factor and updating coefficient, and then the moving object is obtained by using difference method. Finally, the fast local threshold method is used to segment the image to get the target binary image. Because of its high application value and good development prospects. In this paper, the imaging model is established, the corresponding projection matrix is derived, and the camera parameters are obtained by using the three-point calibration method. The experimental results show that the improved inter-frame differential method and the spatio-temporal joint based target detection method are also effective for slow target detection. And the detection method based on space-time joint has better robustness and more outstanding superiority. The visual localization algorithm in this paper is simple to realize and the error is small. The experiment proves the effectiveness of this method.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】