基于车辆行程数据的道路特征识别

发布时间：2018-01-11 06:18

本文关键词：基于车辆行程数据的道路特征识别　出处：《华南理工大学》2015年硕士论文　论文类型：学位论文

【摘要】：蜿蜒起伏的道路形状是影响车辆燃油消耗的重要因素之一,主要体现在两个方面:一是车辆在起伏的道路上行驶时动能与势能之间的转换需要消耗较多的能量;二是当道路的形状特征发生变化时不采取合适的驾驶操作会导致高油耗损失,最高能占据平均燃油消耗的30%。可见在基于道路形状数据的基础上可以实现节能运输线路规划、节能驾驶提醒等改善物流车辆油耗水平的应用,然获得道路的形状数据是一个复杂的系统工程,本文研究如何智能化获取道路形状数据的问题。道路特征包括位置、拓扑结构及形状特征,拓扑结构描述道路网络的连通关系,形状特征包括平面特征及纵面特征,在平面特征包括左弯路、直路、右弯路,纵面特征包括上坡路、平路、下坡路。目前道路特征识别研究集中在拓扑结构的获取,主要有人工测量、影像识别及行程数据推断三种,而对形状特征的获取研究极少,主要采用人工测量的方式,这种方式成本极高,且数据更新速度缓慢。因此本文提出采用行程数据对形状特征进行识别的方案:在正常的驾驶操作中,驾驶员根据实际的道路形状而采取相应的驾驶操作,可见车辆行程数据中隐含着道路形状信息,因此可以通过从行程大数据中构建合适的属性集,然后根据属性集对道路特征进行反向推断。本文的主要工作包括构建合适的属性集合及对道路特征识别问题进行建模。在本文中通过两种方式构建属性集,一是以经过同一路段的多次行程的数据项的统计数据作为属性集,称为统计属性;二是在行程数据的基础上首先识别出驾驶操作,以驾驶操作分布数据作为属性集,称为操作属性。道路形状识别是在已知道路位置及拓扑数据的基础上对形状进行分类或者标注:把该问题建模为分类及序列标注模型,然后采用加权K最近邻、决策树、朴素贝叶斯、隐马尔科夫模型进行求解,实验结果表明平面特征的准确识别率可高达99%以上,而纵面特征的准确识别率可高达90%以上,此外还发现以操作分布作为后续的分类或标注模型的输入属性要优于直接对原始数据项的统计特征。
[Abstract]:The winding road shape is one of the important factors that affect vehicle fuel consumption, mainly reflected in two aspects: one is the vehicle on the up and down conversion between kinetic energy and potential energy consumes more energy; two is the shape feature when the road changes do not take appropriate driving operation leads to high fuel consumption the highest loss, can occupy the average fuel consumption of 30%. visible based on road shape data can be achieved on the energy transport line planning, energy saving driving reminder application of improved logistics vehicle fuel consumption level, so the road shape data obtained is a complex system engineering, this paper studies how to obtain intelligent road shape data. The road features including location, topology and shape features, describe the topology connectivity of road network, including plane feature and shape feature profile Features in the plane features include left bend road, straight, right detours, vertical features including uphill, downhill road, road recognition. Current research focused on the topology acquisition, mainly artificial measurement, image recognition and travel data from three, and the shape features of the acquisition of few studies, mainly by the manual measurement method, the high cost, and the data update speed is slow. Therefore this paper proposes a recognition method using shape features of the travel data: in the normal operation of the driver, the driver to take corresponding driving operation according to the actual road shape, visible vehicle travel data implies the road shape information, so you can by constructing appropriate attribute from the stroke data set, then according to the attribute set of reverse inference on the road features. The main work of this paper includes the construction of appropriate attribute set and Modeling of road feature recognition. Construct attribute set by two methods in this paper, one is the statistical data after several stroke of the same section of the data item as the attribute set, called statistical properties; two is based on data on the first trip to identify the driving operation, the driving operation data distribution as a set of attributes, called operation attributes. Road shape recognition is based on the known road location and topological data on the shape classification or mark: the problem of modeling for classification and sequence labeling model, and then use the weighted K nearest neighbor, decision tree, Naive Bayesian, solve the hidden Markov model, the experimental results show that accurate the recognition of planar feature rate can be as high as 99%, and the accurate identification of characteristics of vertical rate can reach more than 90%, in addition to the operation of distribution as the following classification or annotation model input Properties are superior to the statistical characteristics of the original data items directly.

【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41

【参考文献】