高清自动聚焦及其在网络一体机上的应用研究

发布时间：2018-04-27 20:50

本文选题：自动聚焦 + 清晰度评价　；参考：《浙江大学》2017年硕士论文

【摘要】：近年来,随着数字和网络的发展,网络摄像机已成为安防监控系统中重要的终端组成部分。高清网络摄像机(HD_IPC)相对于传统的模拟摄像机具有高清晰度、高交互性、智能化的特点。因此,高清网络摄像机被广泛地应用于各种监控场合,如银行、安检、公路等,已成为安防行业重要的终端产品。在网络摄像机中,一体化的网络摄像机占据了很重要的一个部分,能够满足不同场景下的高监控需求。自动聚焦技术是一体机中的一项核心技术,对于最终的成像质量有直接的影响。本文研究了高清自动聚焦及其在网络一体机中的应用,主要包括三个核心子算法:图像清晰度评价函数、峰值搜索算法和变焦跟踪算法。本文总结了清晰度评价函数的定性评判标准和定量评判标准,对现有、常用的清晰度评价函数进行总结、分类。本文重点讲述了基于小波分析的图像清晰度评价函数,然后对其进行改进和优化,提出了基于小波变换和梯度函数融合的图像清晰度评价函数,并使用提升小波的方式来减少计算复杂度。本文提出的评价函数相对于其他的评价函数具有更好的'灵敏度和抗干扰性,在噪声环境下也能保持良好的评价性能。本文根据聚焦曲线的理想特性,在现有峰值搜索算法的基础上提出基于新判据的变步长峰值搜索算法。该方法将聚焦曲线的不同部分划分为初始化(Initial)、粗调(Coarse)、中调(Mid)、细调(Fine)4个状态,不同状态有不同的电机步长,在保持准确性的前提下提升了峰值搜索算法的运行速度。本文研究了现有的变焦跟踪算法,对反馈变焦跟踪算法(FZT)进行改进,提出了IFZT算法。IFZT相对于FZT算法,修改了反馈修正点的修正判据,针对大幅离焦状态做了特殊处理,去除了相对复杂的PID算法,相对于FZT方法更加简单、可靠,相比于GZT、AZT方法有更高的跟踪精度。最后本文搭建了以海思Hi3516A为核心的IPC实验平台,并且编写了后端服务程序,进行相应的运行和测试实验。实验结果表明,该实验平台能准确、稳定地实现自动聚焦,满足设计要求。
[Abstract]:In recent years, with the development of digital and network, network camera has become an important terminal component of security monitoring system. HDI PC has the characteristics of high definition, high interactivity and intelligence compared with the traditional analog camera. Therefore, high-definition network cameras are widely used in various monitoring occasions, such as banks, security inspection, highways and so on, which has become an important end product of security industry. In the network camera, the integrated network camera occupies a very important part, which can meet the needs of high monitoring in different scenes. Automatic focusing is one of the core technologies in an integrated computer, which has a direct impact on the final imaging quality. In this paper, the high-definition auto-focus and its application in the integrated network are studied, which includes three core sub-algorithms: image definition evaluation function, peak search algorithm and zoom tracking algorithm. This paper summarizes the qualitative evaluation criteria and quantitative evaluation criteria of definition evaluation function, summarizes and classifies the existing and commonly used definition evaluation functions. This paper focuses on the image definition evaluation function based on wavelet analysis, then improves and optimizes it, and puts forward the image definition evaluation function based on the fusion of wavelet transform and gradient function. The lifting wavelet is used to reduce the computational complexity. Compared with other evaluation functions, the evaluation function presented in this paper has better sensitivity and anti-interference, and can also maintain good evaluation performance in noisy environment. Based on the ideal characteristics of the focusing curve and the existing peak search algorithms, a new criterion based variable step size peak search algorithm is proposed in this paper. In this method, the different parts of the focusing curve are divided into four states: initialized initialer, coarseau, midway, fine Fine.There are different motor step sizes in different states, and the running speed of the peak search algorithm is improved on the premise of maintaining accuracy. In this paper, the existing zoom tracking algorithm is studied, the feedback zoom tracking algorithm is improved, and the IFZT algorithm is proposed relative to the FZT algorithm, the correction criterion of the feedback correction point is modified, and the special treatment for the large defocus state is made. Compared with the FZT method, the PID algorithm is simpler and more reliable, and has a higher tracking accuracy than the FZT algorithm. At last, the IPC experiment platform based on Hayes Hi3516A is built, and the back-end service program is written to run and test the experiment. The experimental results show that the platform can realize auto-focusing accurately and stably and meet the design requirements.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN948.41

【参考文献】