Inferring Road Networks And Mining Human Mobility Based On Private Car Trajectroy Dataset

With the large number of GPS devices installed and used in vehicles,massive vehicle trajectory data is easily obtained.These trajectory datasets potentially reflect urban road networks,human travel behavior,and urban traffic dynamics,etc.,attracting a large number of researchers to use various data mining techniques to try to extract valuable information from these data,then realizing many innovative applications for urban planning,intelligent transportation,environmental protection,and limiting the spread of biological viruses.Based on the private car trajectory data set,this paper proposes a new road network inferring mechanism,based on which the individual travel pattern,the overall travel behaviour of the crowd,and the interaction between cities are explored and analyzed.Road network is an essential part of navigation,assisted driving and autonomous driving technologies.The vehicles' GPS trajectory dataset can reflect the underlying road network.Automatic road networks inferring algorithms directly infer road networks from the collected vehicular trajectory data,which have been attracting a great deal of research works in road networks refining,updating,and building area because of its economies,timeliness,and automaticity.In addition,this way of inferring the potential road network from a large number of trajectories makes the inferred road network itself carry some traffic information,such as the degree of congestion of a certain road segment at certain times,the probability of turning at intersections etc.However,inferring road networks from a large number of vehicle GPS trajectories is also facing some technical challenges due to the low sampling rate of trajectory dataset and the more and more complicated urban road networs,especially the emergence of various complex intersections.This paper attempts to propose a new road network inference method based on the sparsely sampled private car trajectory data set,using the characteristics of the road network itself and the characteristics of the trajectory data set,which can effectively improve the inference accuracy of urban road networks with complex intersections.The private car trajectory dataset can also reflect the long-term mobility of private car owners.This paper analyzes long-term trajectory data of individuals driving private cars,attempts to mine individual travel patterns and quantify the individual travel regularity.In addition,this article also conducts an overall analysis of the mobility of a large number of private car owners to study the characteristics of group travel behaviors,including the stay information analysis at residence,work and leisure places,the distribution of the number of daily trips and daily travel patterns.And the differences in the distribution of travel distance and duration of stay are analyzed and compared,according to the group travel behaviors of the five cities in the Pearl River Delta.Finally,the number of vehicles and the number of activities that the private cars reached among cities reflected the mobility of people among cities,potentially reflecting the differences and integration of social life and economic activities.Based on the private car trajectory data set,this paper proposes a new method for measuring the regularity of individual travel behaviors to explore individual travel patterns.A rule-based algorithm is proposed to effectively identify the place of residence,work and leisure of individuals,so as to analyze the mobility of urban groups in a more granular manner.Finally,two indexes of urban attractiveness and introversion are proposed to analyze the motivation of interaction mobility between cities.This research is based on the private vehicle trajectory data set provided by Baidu and Gaode's strategic partner-Shenzhen Maigu Technology Co.,Ltd.The contributions of this paper are summarized as follows:(1)The more and more complicated road network in cities makes it difficult to infer road networks from city-scale GPS trajectory dataset with low sampling rates.Most of the existing researches are not suitable for inferring road networks with complex intersections.This paper proposes a simplified method to detect intersections,which can eliminate GPS trajectory point interference around complex intersections,so as to accurately identify the location of complex intersections and the direction information of connected road segment.The proposed road networks inferring algorithm uses the direction information of the source intersection to extract all trajectory points on the road segment in a sliding window manner,and finally finds the direct neighbor intersection and the corresponding road segment geometry information.Extensive evaluations demonstrate the effectiveness of the proposed method.The constructed road network matches close to the one from a public editing map Open Street Map,especially the location of the road intersections and road segments,which achieves 92.2% intersections within 20 m and 91.6% road segments within 8m.(2)Private cars have become the main body of urban vehicles,and their mobility can reflect individual travel patterns and regularity.It is a challenge to model individual mobility due to the freedom and diversity of private car trips and the heterogeneity of individual activities.This paper proposes an individual mobility regularity measurement method based on trajectories of private cars,which can help to retrieve various types of travel patterns,and is also useful as a quantitative index to various predicting algorithm,such as dwell time and destinations predictions.Firstly,the individual mobility regularity measurement method retrieves the stop-and-wait information of private cars,which are likely to correspond to the frequently-visit places(FVPs).Secondly,the DBSCAN method is used to detect and identify the FVPs.After that,the individual mobility regularity measurement method leverage Markov chain to elucidate the spatio-temporal transition characteristics when the private cars travel among their FVPs.Then deriving the spatial-temporal entropy rate to conduct a quantitative study for measuring the regularity of each individual private car's mobility.Experiments show that after sorting by entropy rate,the visualization of individuals mobilities shows the transition from regular to irregular,and the proposed regularity measurement method is also more accurate and comprehensive than daily main pattern frequency measurement.(3)Large number of individual travel behaviors can reflect the overall mobility of urban populations.This paper proposes a private car activity region recognition algorithm based on the DBSCAN clustering algorithm.The region of private car activity is effectively classified into three categories: home,work,and leisure.Firstly,this paper uses the inferred road network to help eliminate invalid stops on the road,that is,the traffic jams,to improve the accuracy of the original and destination locations of individual activities.Secondly,the valid original and destination locations of individual activities are clustered and the effective region are identified based on the spatio-temporal relationship.Thirdly,analyze the distribution of the duration of stay,and the start and end of the stay in the places of residence,work and leisure.Analyze the distribution of people's daily trips,distances,and daily patterns.Finally,the overall mobility of the five cities in the Pearl River Delta is compared and analyzed.The comparison of the distribution of travel behavior characteristics of 5 cities in 3 types of regions shows the diversity of people's lives today.Leisure life occupies a dominant position in people's travel.Statistics from daily travel patterns show that although there are many types of daily travel patterns,most people have a certain regularity of daily travel patterns most of the time.(4)Analysis of the trajectory of private cars in five cities in the Pearl River Delta revealed that private car trips are not limited to one city.There are interactions among five cities,which reflects the communication and integration of the inter-city crowd.This paper proposes two indicators of urban attractiveness and introversion to evaluate the motivation of the interaction among cities.It is found that Guangzhou has the most attractiveness to the other four cities,while other cities show more obvious attraction only to neighbor cites.That reflects Guangzhou is a central city in the Pearl River Delta urban agglomeration.The attractiveness of Zhongshan is the weakest,and its introversion is the strongest,and its interaction with other cities is not obvious,then it has a certain degree of independence.