Understanding vehicles commuting pattern based on license plate recognition data

Abstract

Abstract Commuting vehicles are one of the most important vehicles on the road network, especially during morning and evening peak hours. Analysis of commuting vehicles can lay foundation for policy formulation, urban planning, business decision-making. A large number of license plate recognition (LPR) detectors installed on the road network have accumulated a large amount of LPR data that contain the spatio-temporal information on vehicles. Therefore, using the LPR data, the commuting behavior of vehicles can be mined and analyzed. This study takes the LPR data of Hangzhou, China as an example, a series of algorithms are used to extract a total of nine features reflecting the commuting travel behavior using the one-month LPR data, and then the factor analysis method is used to integrate the nine extracted features into three factors. The commuting behavior is characterized from three perspectives: the stability of the vehicle travel behavior during the high frequency travel time period on working days, the stability of the vehicle travel behavior during the non high frequency travel time period, the stability of the origin and destination (OD) of a vehicle during the high frequency travel time period. Based on these three factors, three methods, including the density-based spatial clustering of applications with noise algorithm (DBSCAN), the iterative self-organizing data analysis technique algorithm (ISODATA), and the fast search and find of density peaks clustering algorithm, are used to identify commuting pattern vehicles and their results are compared. Finally, the decision tree algorithm is used to extract the commuting rule and identify commuting vehicles from the collected data. After that, The influence of different high frequency travel time period on commuting rule extraction and commuting pattern vehicle recognition is carefully analyzed. The influence of randomness on the results is also tested. The proposed commuting pattern vehicle recognition algorithm is proved to be robust. The detection frequency and the first and last detected times of commuting vehicles and non commuting vehicles are then analyzed. 78.0% of commuting vehicles were first detected between 06:30 AM and 10:30 AM on each workday on average. And 77.8% of commuting vehicles were last detected between 03:30 PM and 09:30 PM on each workday on average. However, for non commuting vehicles, the phenomenon that start or end the travel during the high frequency travel time period is not obvious. This study lays a foundation for travel policy formulation. For instance, it can provide a reference for the determination of the restricted periods when implementing the travel restriction policy. In addition, the identification of commuting vehicles is extremely useful for public transit network design and optimization, and it can help the decision-making process of a customized commuter bus. In addition, these strategies are expected to reduce the use of private cars and encourage travelers to switch to public transportation.