基于视觉注意的智能视频监控系统的设计与实现

发布时间：2018-03-16 06:20

本文选题：智能视频监控　切入点：视觉注意　出处：《长春工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：视觉注意是人类处理视觉信息的一种方法,而智能监控系统要实现的目标是让摄像头充当我们的眼睛使得我们的生活更智能化,因此将视觉注意机制的算法引入设备中,这使得嵌入式设备更像人的眼睛,并且也能帮助嵌入式设备筛选出重要的信息,加快视觉信息的处理速度。同时,将嵌入式设备接入网络,可以形成庞大的视联网,为城市的各行各业服务,实现真正的智慧城市。本文主要对视觉显著性计算方法与一般的运动目标检测方法进行深入研究,提出将视觉显著性计算方法引入运动目标检测算法中,来检测显著的运动目标,然后结合多目标跟踪实现车流量检测的算法,最后将算法移植到在Exynos 4412处理器平台上。针对本文研究的主要内容,首先对当前主流的视觉显著性计算方法进行研究,并结合算法的对比效果与实时性,选取HC算法计算图像的显著性;然后研究了经典运动目标检测算法中的背景差分法,选取ViBe算法计算运动区域,并对ViBe算法进行了一些改进,可以消除鬼影,并且当出现误检时,可以减少消除误检所需的时间;然后本文提出了一种基于显著图和Vi Be算法的显著运动目标检测算法,该方法利用改进的ViBe算法计算时间显著图和HC算法计算空间显著图,然后以时间显著图中的运动目标为指导,在空间显著图中计算该目标的显著性,并判断该目标是否是显著目标,如果是,则对该目标进行空时显著性融合,得到空时显著图;最后对图像进行自适应阈值分割,获得最终的运动目标,该算法与经典运动目标算法比较,可以有效解决算法中产生的鬼影、阴影以及一些不显著运动目标问题;然后对检测的显著运动进行跟踪,并通过绊线检测来统计各个方向的车流量实现车流量检测,本文多目标跟踪算法使用Kalman滤波与Hungary任务分配算法实现;最后在搭建好的嵌入式软硬件平台上,移植车流量检测算法,实现了智能监控系统中车流量检测的功能。
[Abstract]:Visual attention is a way for human beings to process visual information, and the goal of intelligent monitoring system is to let the camera act as our eyes to make our life more intelligent, so the algorithm of visual attention mechanism is introduced into the device. This makes embedded devices look more like human eyes, and can also help embedded devices to screen out important information and speed up the processing of visual information. At the same time, connecting embedded devices to the network can form a huge visual network. In this paper, the visual salience calculation method and the general moving target detection method are studied deeply. In this paper, the visual salience algorithm is introduced into the moving target detection algorithm to detect the significant moving target, and then the multi-target tracking algorithm is used to realize the vehicle flow detection. Finally, the algorithm is transplanted to the Exynos 4412 processor platform. In view of the main content of this paper, the current mainstream visual salience calculation method is studied, and the contrast effect and real-time performance of the algorithm are combined. Select HC algorithm to calculate the salience of image, then study the background difference method in the classical moving target detection algorithm, select the ViBe algorithm to calculate the moving region, and make some improvements to the ViBe algorithm, which can eliminate the ghost image. And when false detection occurs, it can reduce the time required to eliminate false detection. Then, a salient moving target detection algorithm based on saliency graph and Vi be algorithm is proposed. The improved ViBe algorithm is used to calculate the temporal salience map and the HC algorithm to calculate the spatial salience map. Then the significance of the target is calculated in the spatial salience map under the guidance of the moving object in the time-significant map. If it is, the spatial-temporal salient map is obtained by space-time significant fusion. Finally, the image is segmented by adaptive threshold, and the final moving target is obtained. Compared with the classical moving target algorithm, the algorithm can effectively solve the problems of ghost, shadow and some unremarkable moving objects, and then track the detected salient motion. The multi-target tracking algorithm is implemented by Kalman filter and Hungary task allocation algorithm. Finally, on the embedded hardware and software platform, the multi-target tracking algorithm is implemented on the embedded hardware and software platform. The traffic flow detection algorithm is transplanted to realize the function of the traffic flow detection in the intelligent monitoring system.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN948.6

【相似文献】