基于嵌入式智能相机的目标检测与覆盖方法研究

发布时间：2018-05-27 03:07

本文选题：智能视频监控 + 嵌入式智能相机　；参考：《上海交通大学》2012年硕士论文

【摘要】：随着安防系统在公共治安和预防恐怖袭击等方面的大量应用,人们对大规模智能视频监控系统的需求日益增长。目前基于集中处理的智能视频监控系统由于成本、处理能力、存储能力等因素的限制,无法大规模的使用。嵌入式智能相机具有低成本、低功耗、易扩展的特点,因此嵌入式智能相机网络成为未来大规模监控系统的主要发展方向。本文重点研究基于ARM的嵌入式智能相机的系统搭建、检测算法和覆盖问题,主要工作有如下三个方面。一、嵌入式智能相机平台搭建。在分析嵌入式智能相机特点的基础上,提出新型的高性能嵌入式智能相机设计方案。该方案采用基于ARM9处理器的PXA270开发板作为硬件系统,并开发了多个满足视频监控基本任务的软件模块。二、基于嵌入式智能相机的目标检测算法设计。针对嵌入式智能相机有限的计算和存储能力,提出基于轮廓判定与分类的轻量级目标检测方法。该方法采用“加性增、乘性减”(AIMD)的方法进行轮廓检测,然后对检测到的运动区域进行分类,从而实现前景的辨认和背景的更新。实验证明,所提方法具有速度快、鲁棒性强的特点。三、多相机的优化覆盖策略。由于单个相机的视野限制,监控场景往往要求被多个相机共同覆盖。为了降低监控覆盖成本,结合监控目标的异构需求,提出基于概率的相机优化覆盖策略。该方法充分利用智能相机所具有的PTZ性能,计算各个位置的覆盖概率,在满足各个监控位置的不同监控需求的基础上,有效地降低了多相机条件下的监控成本。最后通过仿真验证了所提方法的有效性。
[Abstract]:With the extensive application of security system in public security and preventing terrorist attacks, the demand for large-scale intelligent video surveillance system is increasing day by day. At present, intelligent video surveillance system based on centralized processing can not be used on a large scale because of the limitation of cost, processing capacity, storage capacity and so on. The embedded intelligent camera has the characteristics of low cost, low power consumption and easy to expand, so the embedded intelligent camera network will become the main development direction of the large-scale monitoring system in the future. This paper focuses on the system construction, detection algorithm and coverage of embedded intelligent camera based on ARM. The main work is as follows. First, the embedded intelligent camera platform is built. Based on the analysis of the characteristics of embedded intelligent camera, a new design scheme of high performance embedded intelligent camera is proposed. The PXA270 development board based on ARM9 processor is used as the hardware system, and several software modules to meet the basic tasks of video surveillance are developed. Second, the design of target detection algorithm based on embedded intelligent camera. Aiming at the limited computing and storage ability of embedded intelligent camera, a lightweight target detection method based on contour determination and classification is proposed. In this method, "additive increase, multiplicative subtraction" (AIMD) is used for contour detection, and then the detected motion regions are classified to realize foreground recognition and background updating. Experimental results show that the proposed method is fast and robust. Third, the optimal coverage strategy for multiple cameras. As a result of the single camera's visual field limit, the monitoring scene often needs to be covered by multiple cameras. In order to reduce the cost of monitoring coverage, a probabilistic camera coverage optimization strategy is proposed to meet the heterogeneous requirements of monitoring targets. This method makes full use of the PTZ performance of the intelligent camera, calculates the coverage probability of each location, and effectively reduces the monitoring cost under the condition of multiple cameras on the basis of satisfying the different monitoring requirements of each monitoring position. Finally, the effectiveness of the proposed method is verified by simulation.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP277;TP368.1

【引证文献】