基于多传感器的特定道路信息识别算法研究

发布时间：2018-03-15 18:23

本文选题：人工智能　切入点：模式识别　出处：《哈尔滨工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着人们生活水平的提升和汽车总体造价的下降,汽车在人们的生活中扮演了越来越重要的角色。车辆在复杂道路状况下的安全驾驶是一个重要的研究方向,道路信息检测则是其中的关键内容。然而,目前对道路信息检测研究较少,特别是车辆作为节点主动探测道路信息的研究较少。在此背景下,本课题研究了基于多传感器的特定道路信息识别算法,研究内容对保证安全驾驶起到促进作用。路面行驶质量指数中最重要的评价标准就是道路的颠簸程度和抗滑性。道路的颠簸程度就是道路的平整度,而抗滑性能评价标准则建立在不同道路类型的基础之上。本课题的研究内容是以汽车为主体的主动探测道路信息的识别问题。由此,本课题选取了最重要的两种信息作为识别目标,一个是路面的颠簸程度信息,另一个是道路的类型信息。对于道路的颠簸类型判断采用了模式识别的思想,研究采用运动传感器获取垂直于路面的加速度,通过特征提取获取能准确描述路面颠簸的类型的特征向量。通过一系列实验验证和参数调优,发现运动传感器结合特征提取与隐马尔可夫模型能较好的判决出道路的颠簸类型。对于道路类型的判断上,同样需要将道路类型具体到能准确描述其信息的物理参数上。现阶段较为成熟的识别物体的方案一般都是建立在机器视觉的基础上,通过一系列实验验证了在道路类型的识别上,纹理特征能较好的表示道路的类型,最终选取的纹理特征算法为灰度共生矩阵法。同时分类器的选取方面,为了解决道路类型识别时的遮盖问题,文中提出了投票式支持向量机的应用方法,并提出了完整的道路类型识别方案。论文最后为了验证文中提出算法与应用方法。在道路颠簸识别方面,搭建了相应的实验平台。利用这个平台实时采集车辆垂直方向的加速度数据,并通过特征提取和分类仿真得到了满意的效果,平均识别精确度可以达到94%。在道路类型识别方面应用仿真,验证了相应的算法和应用。实验验证,以计算机视觉为基础的纹理特征搭配改进的支持向量机算法能较好的识别道路类型,精度上能达到较好的效果,平均识别精确度达到93.2%。总体来说文章中所使用的算法和使用方法对于保证安全驾驶有较高的实用价值和使用价值。
[Abstract]:With the improvement of people's living standard and the decline of the overall cost of automobile, automobile plays a more and more important role in people's life. The safe driving of vehicles in complex roads is an important research direction. Road information detection is one of the key contents. However, there are few researches on road information detection, especially on vehicles as nodes to detect road information. In this paper, a multi-sensor based road information recognition algorithm is studied. The most important evaluation criteria in road driving quality index are the bumping degree and skid resistance of the road. The bumpy degree of the road is the smoothness of the road. However, the evaluation standard of anti-skid performance is based on different road types. The research content of this subject is the identification of road information with the automobile as the main body. In this paper, the most important two kinds of information are selected as the recognition target, one is the road bumping degree information, the other is the road type information. In this paper, the acceleration perpendicular to the road surface is obtained by motion sensor, and the eigenvector which can accurately describe the type of road bumps is obtained by feature extraction, which is verified by a series of experiments and optimized by parameters. It is found that the motion sensor combined with feature extraction and hidden Markov model can better judge the type of road turbulence. It is also necessary to specify the type of road to the physical parameters that can accurately describe its information. At the present stage, more mature schemes for identifying objects are generally based on machine vision. Through a series of experiments, it is proved that the texture feature can represent the road type well in road type recognition, and the final texture feature algorithm is gray-scale co-occurrence matrix method. At the same time, the selection of classifier is also discussed. In order to solve the covering problem of road type recognition, this paper presents an application method of voting support vector machine (VSVM). Finally, in order to verify the algorithm and application method in this paper, in the aspect of road bumping recognition, the paper puts forward a complete road type recognition scheme. This platform is used to collect the acceleration data of the vehicle in the vertical direction in real time, and the result is satisfactory through feature extraction and classification simulation. The average recognition accuracy can reach 94 points. The simulation is applied to road type recognition, and the corresponding algorithm and application are verified. Texture features based on computer vision and improved support vector machine (SVM) algorithm can recognize road types well and achieve better accuracy. The average recognition accuracy is 93. 2. In general, the algorithms and methods used in this paper have high practical value and use value to ensure safe driving.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;TP212

【参考文献】