Road Feature Recognition Based On The Vehicle Travel Data

The rough road is one of the important factors affect vehicle fuel consumption.It is reflected in two aspects: first, the conversion between the kinetic and potential energy need more energy when driving on the rough roads;second, in the driving process,the bad driving operating occurred mainly in the road of that the shape of the feature changes,the driver does not take suitable appropriates in time, this can causes the energy loss, this part of loss can as much as the fuel consumption of about 30%.So, basic on the road shape data,It can achieve some energy save system such as transportation route planning system, eco-driving reminders system and so on to improve logistics vehicle’s economic performance. However, getting the road shape data is a complicated systematic project. This paper will study the problem of how to getting the road shape data in a intelligent way.The road characteristics include location,topology and shape features.The topology describe the connectivity of network,which has a wide application in the major map providers currently.Shape features including flat and vertical features,the flat features divided into left turn,straight and right turn, the vertical features divided into uphill,planar and downhill.With the increasing of the road network,using traditional way such as field measurement will whith a high cost and much time.Thus this paper presents a program that make road feature recognition based on vehicle travel data.In driving,the driver take appropriate driving operation according to the actual road shape, so the vehicle trip data implies road shape information, it is possible to constructing a suitable set of attributes from travel data and then using the attributes set to do road feature recognition.The main works in this paper includes constructing a suitable set of attributes, make the road feature recognition model and solving this model.This paper using two ways to build the attributes set.The first way is to statistics data item group by the same road,it called the statistical attributes. In the second way the driving operation will be identify first,and then use the distribution of driving operation as attributes,which called the operational attributes.The road feature recognition problem is under conditions of the existing road location and topology data, it is modeled as a classification and sequence labeling model.The weighted K-nearest neighbor,decision trees,naive Bayes and hidden Markov model are used to this problem.Experimental results show that it can get a accuracy up to 99% in flat feature recognition and a accuracy up to 90% in vertical feature recognition, the operational attributes is better than the statistical attributes.