基于信号分解表示的交通标志定位与识别算法研究

发布时间：2018-04-24 07:53

本文选题：稀疏表示 + 非负矩阵分解　；参考：《大连理工大学》2015年硕士论文

【摘要】：随着社会的快速进步和经济的高速发展,一、二线城市的机动车数量成爆炸性逐年增长,从而不可避免地产生了包括交通拥堵等负面影响,驾驶员如何安全地驾驶也引起了普遍关注,智能交通应运而生。而在智能交通系统所涉及的许多计算机视觉(Computer vision)技术领域当中,交通标志的定位与识别又是极其重要的组成部分。交通标志的定位与识别系统目的是在机动车行驶过程中,快速地搜索交通标志然后正确地获取交通标志携带的主要信息。因为该领域具相当高的实用价值,即提高了驾驶的安全性,因此,多年来一直是学者们一个重要的研究课题。在阅读了大量相关论文和其他领域参考文献后,本文将新的算法应用于交通标志的定位与识别系统。本文的创新点包括：(1)增加定位阶段的自适应性,可以在大部分不同天气条件下正确定位；(2)基于稀疏分解表示算法,设计训练了级联形式的字典,并基于该字典实现对交通标志的稀疏分解,通过分解系数完成交通标志的识别；(3)将非负矩阵分解的改进形式用于图像分类。每次迭代时保持字典W不更新,只更新系数矩阵H,相比稀疏表示算法增加了各部分的物理意义。本文通过对实际拍摄的包含交通标志的图像进行标志的定位,另外使用德国交通标志(GTSRB)提供的40余种标志库进行训练和识别测试,并且与已有算法各性能和特点相比较,结果表明本论文提出的基于稀疏表示和非负矩阵分解的两种创新方法,平衡了实时性和识别率,同时对于光照,旋转和遮挡的鲁棒性有一定的提升。
[Abstract]:With the rapid progress of the society and the rapid development of the economy, the number of motor vehicles in the first and second tier cities has been explosively increasing year by year, which inevitably has negative effects, including traffic congestion and so on. How to drive safely has also aroused widespread concern, and intelligent transportation has emerged as the times require. The location and recognition of traffic signs is an extremely important part in many fields of computer vision technology involved in Intelligent Transportation system (its). The purpose of the traffic sign location and recognition system is to search the traffic sign quickly and get the main information of the traffic sign correctly. Because this field has a high practical value, that is, to improve the safety of driving, it has been an important research topic for many years. After reading a large number of related papers and other references, this paper applies the new algorithm to the location and recognition system of traffic signs. The innovations of this paper include: (1) increasing the adaptability of the positioning stage, which can be correctly located under most different weather conditions.) based on the sparse decomposition representation algorithm, a dictionary in cascaded form is designed and trained. Based on the dictionary, the sparse decomposition of traffic signs is realized, and the improved form of non-negative matrix decomposition is applied to image classification by using the decomposition coefficient to recognize traffic signs. In each iteration, the dictionary W is not updated, only the coefficient matrix H is updated. Compared with the sparse representation algorithm, the physical meaning of each part is increased. In this paper, we use the more than 40 kinds of sign library provided by GTSRB to train and identify the actual images with traffic signs, and compare them with the performance and characteristics of the existing algorithms. The results show that the proposed two innovative methods based on sparse representation and non-negative matrix decomposition balance real-time and recognition rate, and improve the robustness of illumination, rotation and occlusion.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495;TP391.41

【参考文献】