基于数据融合的运动目标识别方法研究

发布时间：2019-01-14 10:51

【摘要】：智能视频监控现已被广泛地运用在安防系统当中,且主要安装在仓库、银行、机场、交通路口等重要区域。因此提高视频监控中运动目标检测识别的准确度成为我们当前研究的重点。本文主要针对如何提高智能监控系统中运动目标的检测识别率问题,对现有的目标检测识别技术进行系统的分析研究,并提出了改进方案,最后利用数据融合方法进一步提高了目标识别的精确度。主要研究内容如下:提出了一种基于自适应惯性权重优化萤火虫算法(IFA)改进的二维OTSU图像阈值分割法,在保证分割效果的前提下提高了二维OTSU阈值分割法的运算效率。在此基础上,针对监测运动目标获得相对完整信息的需求,提出了一种结合基于IFA的二维OTSU阈值分割法改进三帧差法(ITFD)与光流法的运动目标检测方法,使得整个检测能够达到实时性要求的同时取得更好的目标检测效果。围绕运动目标区域检测结果,采用基于被检测目标物理特征以及HOG特征结合支持向量机分类器的目标识别方法完成单目标识别,并通过运用“一对一”的方法实现了多目标分类,从而有效的提高了系统的识别性能。针对单视角目标识别中出现误判等问题,提出了基于多视角图像决策层数据融合技术的目标识别决策融合方法,以获取更加准确的类别判断。综上所述,本文针对运动目标检测识别中的阈值分割、运动目标检测、单目标识别和误判修正等问题提出了改进算法和实现策略,有效地提高了视频监控系统对多运动目标域的单体特征目标的识别精度。实验验证表明,本文提出的有关算法能够准确的检测和识别视频图像序列中的运动目标,较传统的目标检测识别方法准确率和鲁棒性都得到了有效提高,同时基于多视角数据融合的方法可以有效纠正由于单摄像头识别引起的错误判别问题,有效提高了系统的识别准确度。
[Abstract]:Intelligent video surveillance has been widely used in security systems, and mainly installed in warehouses, banks, airports, traffic junctions and other important areas. Therefore, improving the accuracy of moving target detection and recognition in video surveillance has become the focus of our current research. Aiming at the problem of how to improve the detection and recognition rate of moving targets in intelligent monitoring system, this paper makes a systematic analysis and research on the existing target detection and recognition technology, and puts forward an improved scheme. Finally, the accuracy of target recognition is improved by using data fusion method. The main research contents are as follows: an improved threshold segmentation method for two-dimensional OTSU image based on adaptive inertial weight optimization firefly algorithm (IFA) is proposed, which improves the efficiency of two-dimensional OTSU threshold segmentation method on the premise of ensuring the segmentation effect. On this basis, aiming at the need of obtaining relatively complete information for monitoring moving targets, a new method of moving target detection based on 2D OTSU threshold segmentation method based on IFA is proposed, which is improved by three-frame difference method (ITFD) and optical flow method (OFC). So that the whole detection can achieve real-time requirements and achieve better target detection effect. Around the result of moving target region detection, a single target recognition method based on physical feature and HOG feature and support vector machine classifier is adopted. The method of "one-to-one" is used to realize multi-object classification, which improves the recognition performance of the system effectively. Aiming at the problem of misjudgment in single-angle target recognition, a method of target recognition decision fusion based on multi-view image decision level data fusion technology is proposed in order to obtain more accurate classification judgment. To sum up, this paper proposes an improved algorithm and implementation strategy for threshold segmentation, moving target detection, single target recognition and misjudgment correction in moving target detection and recognition. The recognition accuracy of single feature targets in multi-moving target domain is improved effectively by video surveillance system. Experimental results show that the proposed algorithm can accurately detect and recognize moving targets in video sequences, and the accuracy and robustness of the proposed algorithms are improved effectively. At the same time, the method based on multi-view data fusion can effectively correct the error discrimination problem caused by single camera recognition, and effectively improve the recognition accuracy of the system.
【学位授予单位】：沈阳大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】