【作者】 林鹰；

【导师】 何丕廉；

【作者基本信息】 天津大学 ， 计算机应用技术， 2008， 博士

【摘要】 储备粮实物监管与稽核是我国储备粮管理的核心,但由于实物监管与稽核体系的缺陷,以及监管与稽核的客观困难,使得虚库、虚报补贴、以差换好、偷盗等各类违规、违法事件层出不穷,给国家造成巨大经济损失,削弱了国家宏观调控和社会稳定。本文以省级储备粮监管与稽核为背景,通过粮食实物监管、稽核问题与体系分析,总结了导致弊端的原因和机理,针对体制、机制瑕疵,以及监管和稽核工具的严重落后,以填补监管空白和落实实物稽核思路,提出了以计量工具创新和IT技术应用为基础的智能监测、全程监管、智能稽核、证据留底、据实补贴的全新监管和稽核理念,以及完善监管与稽核体系的对策方案。针对自动监管与稽核的三大技术核心:重量自动测量、远程视频监管及码流管理,进行了技术实现方案的分析与论证。在重量自动测量方面,通过称重计量法、主动测量法、基于图像识别技术的方法三类技术的分析、比较,确定了以单目视觉识别为核心的技术实现路径。重点研究了不同场景的数量识别方法,采用色调结合梯度阈值的二维分割方法有效的解决了边界的确定与闭合问题,再将所分割边界用傅立叶描述子表述,通过其傅立叶描述子和色调参数与先验知识多维特征函数的比对实现对象体的精确分割与高可靠判别,有效解决了散装粮场景和袋装粮场景的粮食数量识别问题。进而结合移动侦测等技术,采用自动布防和首尾帧判别方法,解决了码流动态性判别、批次判别等问题;在远程视频监管方面,对RDP/DP通信机制进行改造,采用无差错检测和码流传送参数回调的方法确定码流段的传输方式与重传机制,确保码流段的完整送达;在码流数据库存储与管理方面,通过识别和提取码流段报警类型、端口地址、时间、变化类型等特征,建立码流段特征与码流段的唯一映射关系,以两级数据库模式实现码流数据的存储与管理,大大提高了码流管理使用的安全性和方便性。最后,设计并开发了基于视频的储备粮智能动态监管稽核系统。系统由视频监控平台系统、视频图像码流数据库管理系统、视频图像动态识别系统、识别数据管理系统、智能监测与报警系统五部分构成。该系统采用单目图像动态识别技术,以实物数量智能检测为核心,实现了粮食实物数量的远程全自动监管和稽核,以及相关证据保留与管理,实物数量计量误差小于3%,满足管理要求,使用方便、可靠,系统实现成本较低。更多还原

【Abstract】 The supervision and audit are the core of our country’s grain reserve management,but because of the drawbacks of our grain reserve’s supervision and audit system,which led to many illegal affairs, such as empty storage, falsely reporting for moresubsidy, changing the good by the inferior and so on, piling up one after another.These irregularities result in huge economy loss for our nation, which influence ournational macroscopically adjustment and societystability.This text takes the provincial grain reserve’s supervision and audit forbackground. It sums up the reasons and mechanism which led to these irregularities,through analyzing the system of our grain reserve supervision and audit. In allusion tothe flaw of the management mechanism and the seriously undeveloping of the audittools, we put forward a new principle for supervision and audit, including theintelligent detection, the whole distance supervision, intelligent audit, evidencesubsistence and subsidizing actually, based on the innovation of calculating tools andthe IT technique applications. We also put forward a countermeasure to perfect thesupervision and audit system, in order to fill up the blank of supervision and carry outthe audit idea.To the three greatest techniques of the supervision and auditor, the auto weighingmeasures, the long range video frequent supervision and proof management, we carryon the analysis and demonstration of its technical implementation. In the aspect ofauto weighing measure, make certain of the way of technical implementation based onmonocular vision, by compare of three technical method, weighing computationmethod, active diagraph method and the method based on image recognition. Thepoint studies the quantity identifies method in different scene, adopting the twodimensional segment method combine hue with threshold value to effectively solvethe problem of boundary confirming and boundary shutting. Then express theboundary by Fourier descriptors. Through the comparison between boundary Fourierdescriptors and hue information with the n-dimensional eigenfunction prioriknowledge, the accurately segment and high reliably recognition were come true,effectively solving the grain quantity’s measure. Furthermore, we combine it with motion detection to solve the problem of the distinction of bit stream’s dynamic andbatch identitification. In the aspect of long range video frequent supervision, weimprove on the RDP/DP protocol, adopting the method of channel bandwidthdetection and callback of bit streams parameter to confirm the way of bit streams’transfer, and mechanism of retransmission, in order for the entire integrallty arrive. Inthe aspect of bit steams’s reserve with database, we combine the way of just reservethe pivotal bit streams with the way of constringent server, saving 99% space. It builtthe unique mapped between bit steam character with a section of bit stream, so as tomake the server and management of bit stream data come tree, through the databasemanagement of bit stream section’s character. The management of bit steam are moresafelyand convenience.Finally, this paper presents a video-based, intelligent dynamic reserved grainsupervision and auditing system about the reserved grain.The system is consist of thevideo surveillance systems by the platform system, video image stream databasemanagement systems, dynamic video image recognition systems, identification datamanagement systems, intelligent monitoring and alarm systems.The system uses a single-head dynamic image recognition technology, at thecore of the number of in-kind intelligence detected, achieved a number of long-rangekind of food automatic monitoring and auditing, and the relevant evidence and toretain management.The physical quantity measurement error is less than 3percent.The systems meet regulatory requirements.It’s easy to use, reliable and lowercost.更多还原