Traveltime Calculation Research Based On Massive Recognition License Plate Data Sets

The calculation of travel time in urban road has been one of the core issues in the study of the intelligent transportation systems. Accurate and efficient calculation of travel time can effectively help us to analyze the urban congestion and vehicle driving conditions. So it is an important basis for urban traffic management and control. The calculation of travel time in intelligent transportation relies on a large number of related traffic data acquired from the urban transport system. The Vehicle License Plate Recognition (VLPR) data, which has the properties of wide coverage, continuous collecting and accurate locating of vehicles, is a new kind of data source in the domain of intelligent transportation. Therefore how to utilize the huge amounts of VLPR data to calculate travel time of urban roads is becoming an important problem in intelligent transportation applications.In order to solve the problem of travel time calculation based on the huge amounts of VLPR data set, this paper conducts the research from two points of view, namely the computing model and implement of distributed computing model, after the analyzing and summarizing of related work and the existing research results. The main work of this paper includes:(1)With continuous growing and large-scale vehicle license plate recognition data collection, the calculation model for actual measurement of travel time was proposed based on VLPR data. Furthermore, a calculation model for the prediction of travel time is also designed by classificating the rate of change of adjacency time interval’s travel time according to the naive Bayes thought.(2)Based on the proposed calculation model for travel time, in order to solve the computing performance issue, we design a parallel computing method for travel time computing based on spatio-temporal data partition in a distributed environment. The corresponding travel time computation system was implement based on the Hadoop opensource platform.(3)A set of experiments based on real and million-level VLPR data sets show that, the computing performance of our method in this paper can be improved by more than three times compared to traditional travel time computing method. Meanwhile our method is more suitable for fine-grained partition, unaffected by the road network scale and good extensibility. In addition, the experiments of travel time prediction can show great accuracy for fine-grained partition, and the accuracy rate is above85%.