视频监控中的遗留物检测技术的应用研究

发布时间：2019-01-23 18:58

【摘要】：智能视频监控系统能够自动地监控场景,当发现场景中有违规行为时立即引发警报,从而大大减小了工作人员的工作量,提高了检测的准确率,所以智能视频监控系统与传统的视频监控系统相比更适用于安保工作。目前放置遗留物品是恐怖袭击的主要手段之一,尤其在那些人口流动比较大的公共场所遗留物体的检测问题更不能被忽略。对于人口密集的公共场所和一些安全级别较高的部门进行实时的、全天候的遗留物体检测就变得特别重要。本文研究的主要内容是基于视频的遗留物检测算法,其主要任务是对监控场景进行遗留物判别和放置者检测。遗留物检测算法主要分为两大类分别是基于跟踪的方法和基于目标检测的方法,本论文中采用的是基于目标检测的遗留物检测算法。在本文中对原算法做出了两个方面的改进。一方面,为了降低计算的复杂度,把算法的前景提取算法改成了混合差分法;另一方面在原算法中加入了对部分静止物体的判断。具体来说,本文遗留物检测算法使用的是基于静止前景物体减法的方法,该算法主要分为背景维护、遗留物体判别、放置者检测三个阶段。在背景维护阶段,采用混合差分法建立背景模型,并且得到前景图像,使用这种方法既可以得到运动前景物体也可以得到静止的前景物体;在遗留物体判别阶段,为前景建立三维模型把前景物体分为运动前景物体和遗留物体。基于混合差分法和三维建模的遗留物检测算法对于移走物体所引起的“鬼影”、物体的部分静止和遮挡等都有较高的鲁棒性。在遗留物放置者检测阶段使用的是视频回溯法,找出遗留物体刚进入该区域和完全进入该区域的一段视频数据,计算出其所有图像中候选放置者所在区域的颜色直方图,然后计算其的平均值,把直方图最接近平均值的那帧图像中的候选放置者视为真正的放置者。此外,详述了实验的运行环境和系统的整体结构,设定了算法中的主要参数值,用不同标准视频库中的数据对该算法进行功能测试和性能评估。实验结果表明该算法能够较好的完成遗留物检测和放置者检测的任务,并且有较好的鲁棒性和准确性。最后,运用该遗留物检测算法设计了一个遗留物体检测系统。
[Abstract]:The intelligent video surveillance system can automatically monitor the scene, and when there are violations in the scene, the alarm can be triggered immediately, which greatly reduces the workload of the staff and improves the accuracy of detection. So the intelligent video surveillance system is more suitable for security than the traditional video surveillance system. At present, the placement of residual objects is one of the main methods of terrorist attacks, especially in those public places where population mobility is relatively large, the detection of residual objects should not be ignored. Real-time, round-the-clock detection of legacy objects is particularly important for densely populated public places and some higher-security departments. The main content of this paper is the video based residue detection algorithm, the main task of which is to detect the residue and the placer of the monitoring scene. Legacy detection algorithms are divided into two main categories, one is based on tracking method and the other is based on target detection. In this paper, the algorithm based on target detection is used. In this paper, the original algorithm is improved in two aspects. On the one hand, in order to reduce the computational complexity, the foreground extraction algorithm is changed into the hybrid difference method; on the other hand, the judgment of some static objects is added to the original algorithm. Specifically, this algorithm is based on static foreground object subtraction, which is divided into three stages: background maintenance, residual object discrimination, and placement detection. In the stage of background maintenance, the background model is established by using the mixed difference method, and the foreground image is obtained. By using this method, the moving foreground object can be obtained as well as the stationary foreground object. In the phase of discrimination, the foreground objects are divided into moving foreground objects and leftover objects. The hybrid difference method and 3D modeling based detection algorithm are robust to the "ghost" caused by the removal of objects, the partial stillness and occlusion of objects. A video traceback method is used in the detection phase of the remnant Placer to find out a segment of video data that has just entered the region and enter the region completely, and to calculate the color histogram of the region in which the candidate Placer is located in all of its images. Then the average value is calculated and the candidate Placer in the frame of the histogram closest to the average is regarded as the true Placer. In addition, the running environment of the experiment and the whole structure of the system are described in detail, the main parameters of the algorithm are set up, and the function and performance of the algorithm are tested and evaluated with the data from different standard video libraries. The experimental results show that the proposed algorithm can accomplish the task of detecting the remnants and the placer, and has good robustness and accuracy. Finally, a legacy object detection system is designed by using the algorithm.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;TN948.6

【相似文献】