蔬菜大棚监控视频处理关键技术的研究

发布时间：2018-09-09 14:13

【摘要】：随着我国农业经济的发展,人们对农产品从量的需求转向为质的需求。如今农产品质量安全事件频发,为了人们的身体健康和良好的农产品市场秩序,建立农产品溯源体系是十分必要的。蔬菜大棚的监控是农产品溯源中的一个重要环节,对蔬菜大棚的生产过程进行实时监控并将监控视频中的有用信息嵌入到溯源体系中,可有效提高农产品质量安全水平。在监控视频中有大量的冗余信息,大部分有效信息都存在于视频的前景中,因此需要进行运动前景与背景分离的工作。目前传统的运动检测算法在较为复杂的场景下的适用性较差,在天津食品安全低碳制造协同创新中心的支持下,本文提出了一种改进的码书模型算法,并与BP神经网络和压缩算法结合,扩展了运动目标检测的应用。本文主要成果包括:1.对比几种传统的运动检测算法的优缺点,通过实验验证选取检测效果较好的码书模型并加以改进,使其具有更好检测效果并能适应更为复杂的场景。2.对码书模型提取出的运动前景做进一步检测,通过采用BP神经网络对前景目标进行分类,得到应用中实际需要的有效前景目标。3.运用视频压缩算法对视频中有前景出现的部分和无前景出现的部分进行不同的压缩处理,在保证信息有效完整的同时减少视频数据量,使其更好地嵌入到农产品溯源系统中。
[Abstract]:With the development of agricultural economy in China, the demand for agricultural products has changed from quantity to quality. Nowadays, the quality and safety of agricultural products occur frequently. It is necessary to establish traceability system of agricultural products for the sake of people's health and good market order of agricultural products. The monitoring of vegetable greenhouse is an important link in the traceability of agricultural products. It can effectively improve the quality and safety of agricultural products by real-time monitoring the production process of vegetable shed and embedding the useful information from the surveillance video into the traceability system. There is a lot of redundant information in the surveillance video, most of the effective information exists in the foreground of the video, so it is necessary to separate the motion foreground from the background. At present, the traditional motion detection algorithms have poor applicability in the more complex scenarios. With the support of Tianjin Food Safety low carbon Manufacturing Collaborative Innovation Center, an improved codebook model algorithm is proposed in this paper. Combined with BP neural network and compression algorithm, the application of moving target detection is extended. The main achievements of this paper include: 1. By comparing the advantages and disadvantages of several traditional motion detection algorithms, the codebook model with better detection effect is selected and improved through experiments, so that it has better detection effect and can adapt to more complex scene. 2. The moving foreground extracted from the codebook model is further detected, and the effective foreground target. 3 is obtained by using BP neural network to classify the foreground target. The video compression algorithm is used to compress the parts with and without foreground in video, which can reduce the amount of video data and embed it into the traceability system of agricultural products.
【学位授予单位】：天津科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】