OpenStreetMap(OSM) has recently emerged as a promising solution to the challenging task of creating an accurate and up-to-date map of the fast changing world. However, as this database is primarily created by amateurs, the quality of its data is unknown unless a comparison to ground truth is performed. Road networks form a significant feature of OSM and in the past several researchers have tried to assess their quality by using techniques which usually involve some sort of referencing to maps captured by the corresponding national mapping agency. However, finding some automated machine learning based solution to solve this problem of quality assurance seems to be a natural choice. Nevertheless, in order to apply any algorithm successfully, an appropriate representation of the underlying data is a prerequisite. The two predominant street network representations (namely, the primal and dual) fail to represent data in a form suitable for machine learning. Therefore, this paper presents a simple and intuitive representation for street networks where information like street name and category is easily available. This representation offers the combined advantages of primal and dual forms and allows for an efficient application of machine learning algorithms towards quality assessment of street networks datsets.
Recent technological trends enable modern traffic prediction and management systems in which the analysis and prediction of movements of objects is essential. To this extent the present paper proposes IncCCFR---a novel, incremental approach for managing, mining, and predicting the incrementally evolving trajectories of moving objects. In addition to reduced mining and storage costs, a key advantage of the incremental approach is its ability to combine multiple temporally relevant mining results from the past to capture temporal and periodic regularities in movement. The approach and its variants are empirically evaluated on a large real-world data set of moving object trajectories, originating from a fleet of taxis, illustrating that detailed closed frequent routes can be efficiently discovered and used for prediction.
Identifying spatial patterns of traffic collisions is critical for improving the efficiency and effectiveness of the deployment of traffic enforcement resources as well as road safety. In recent years, many studies have focused on finding locations with high collision concentration, so-called hotspots, without integrating the likely available non-spatial attributes into analysis. In this paper we propose a method for identifying the sets of non-spatial attribute-value pairs (AVPs) that together contribute significantly to the spatial clustering of the corresponding collisions. We call such a set of AVPs a Spatial Co-Clustering Pattern (SCCP). By applying our method on the city of Edmonton's historical collision data, we discovered a larger number of meaningful hotspot patterns than traditional hotspot analysis methods did, and revealed the relevant non-spatial indicators for explaining those hotspots.
This paper discusses the problem of wayfinding in complex environments. Movement is complicated in many ways, including the existence of obstacles that impede travel across continuous space. To support route planning and navigation for the blind, shipping, robotics, etc., methods have been developed to identify the shortest or most efficient path. This paper details a GIS based method for finding an optimal path for this problem. The developed approach offers substantial computational advances over existing approaches, making it appealing for real time travel support. Application results are presented to demonstrate the effectiveness of the approach.
Many acceleration techniques exist for the single-pair shortest path problem on road networks. Most of them have been significantly improved over the years to achieve faster preprocessing times and superior performance. In this spirit, our current work significantly improves the classic ALT (A* + Landmarks + Triangle equality) algorithm. By carefully optimizing both preprocessing and query phases, we managed to effectively minimize preprocessing time to a few seconds, making the ALT algorithm also suitable for dynamic scenarios, i.e., road networks with changing edge weights due to traffic updates. We also accelerated the query phase for both unidirectional and bidirectional versions of the ALT algorithm, providing fast enough query times (including full-path unpacking) suitable for real-time services and continental road networks.
GPS positioning devices are becoming a commodity sensor platform with the emergence and popularity of smartphones. This abundance of GPS trajectories has fueled significant research around map-matching and related applications such as traffic assessment and prediction. Unfortunately, this research has only been used in costly and complex fleet management solutions. Our latest research endeavor addresses this issue by presenting cost-effective solutions for adapting state-of-the-art research around map-matching and live traffic assessment in the context of fleet management applications. This paper showcases various research results wrapped in a single extensible fleet management platform.
As users increasingly rely on navigation assistance, they expect accurate and reliable navigation performances, especially in obstructed areas where positioning quality of global navigation satellite system (GNSS) is often deteriorated. This paper presents a novel methodology, called integrated GNSS (iGNSS) quality of service (QoS) prediction to provide a means for navigation applications to plan according to GNSS positioning quality. The methodology could be used to predict iGNSS QoS on prospective route segments ahead of time, among other tasks, allowing the navigation service to prepare a suitable plan before the users reaches the poor iGNSS QoS segments. The results were evaluated by comparing the predicted iGNSS QoS to the collected GPS data on sample routes in different environment surroundings under various parameter settings. The results show that the proposed methodology is capable of predicting GNSS QoS on segments under open sky condition accurately and identifying locations on segments with poor GNSS QoS.
This paper provides a sample for acquiring and processing crowd sourced mobile sensor data. An infrastructure for receiving and storing has been developed as well as the corresponding clients that collect smartphone sensor data and send them to the server. Tests and statistics were generated to get first impressions how data logging and storing will work. To analyze the collected data, a web based visualizing toolkit has been connected as well as a processing framework to generate refined geodata. Giving an example on possibilities with crowd sourced sensor data a classification approach using crowd generated categories and data mining methods.
VTIS is a dynamic notification system that takes in a user's route and calculates the time-delay imposed by disruptions to the normal traversal. The disruptions are calculated by using crowdsourced notifications. This is accomplished by the creation of a client side application for notification display and a server infrastructure that will process and store the event information. We have devised a system that will generate personalized notifications for users based on a provided path, temporal range, and set of transportation modes. At a high level, the functionality of this system is to identify events that affect the user's route and notify the user of these events. The VTIS will provide a multimodal notification system based on information mined from Twitter data and volunteered information from VTIS users. This information will be stored to create a repository of transportation events., This repository will be queried to notify affected users of events that may affect their route.. Although this outlined problem has been solved previously, our approach is novel in several ways: (1)accounting for multiple modes, (2)combining user input with mined data, and the (3)modeling method used to calculate effects on the user's route. Some of these methods have been implemented separately; however, a comprehensive system has not been constructed that includes all of these items.
Collaborative transportation has been proposed as a potential solution to decrease congestion, reduce environmental effects of transport, and provide transportation options to those with no or restricted travel options. One such system is demand-responsive transport, in which passengers share a vehicle, usually a bus, but can be picked up or dropped off at a passenger-specified location and time. However, as these systems are expensive to implement and require long trials in order to gain traction, effective simulation is required in order to explore their viability before implementation. Although previous work has concentrated on the number of trip requests, the spatial distribution of these requests has not been considered. This paper explores four spatially-varying demand patterns -- random, a many-to-one scenario, all short distance trips, all long distance trips -- using a simulation of an ad-hoc demand-responsive bus system. It is shown that along with the number of trip requests and the requested trip distances, the spatial distribution of passengers does indeed have an effect on the level of service.
This paper addresses the creation and maintenance of partitions of city surfaces for mapping and transportation applications. It proposes a hierarchical spatial surface partitioning, encoding the spatial partition with a 2D arrangement and structuring a generic hierarchy of semantic objects with a directed acyclic graph (DAG), in which the leaves point to the partition elements (polygonal regions, line strings, points). Semantic objects such as buildings, sidewalks and roads are described by grouping other objects and partition elements with their semantic relationships. In the proposed generic data model, geometry and spatial relationships of the semantic objects are respectively described by the geometry and topology of the planar partition. The proposed geometric data structure for creating and maintaining this partition is a 2D arrangement. In addition, the hierarchical object model encodes the thematic and semantic relationships between the objects. Besides the data model, methods and algorithms are discussed for leveraging existing vector datasets to create and maintain such partitions. These partitions are then fit to further processing and analysis using computational geometry and graph theory algorithms. For this purpose, three application-wise generic algorithms were integrated into our system called Streetmaker: two skeleton operators for centerline generation (straight skeleton and medial axis) and connectivity graphs for itinerary calculations. Moreover, specific algorithms can be integrated into Streetmaker for specific applications. We demonstrated an example usage of this framework for generating static obstacle avoiding pedestrian network graphs. The representation of the network graph and the process used to generate it, can be considered as the second contribution of our work besides the proposed data model.
Navigating multiple first responders to multiple destinations avoiding moving obstacles poses a set of challenges for navigation researchers. When first responders are performing their tasks, new data about the moving obstacles and new tasks might arrive, which create needs for re-planning of routes and re-allocation of tasks. To be able to do so, they need a path planner that is capable of processing large volumes of spatio-temporal data from different resources, and able to generate routes as quickly as possible. In this paper, we propose a multi-agent infrastructure for navigating multiple responders to multiple locations avoiding moving obstacles. To our knowledge, this type of navigation has not yet been addressed in the literature. We design and develop a set of software agents according to their roles in emergency response, which assists emergency actors in data collection, data processing, route generation, and task allocation. We also use an auction-based approach to allow the agent system to allocate different target locations to the responders. We apply our system to a real road network and show some preliminary results.
This paper deals with the classification problem of modes of transportation using the information obtained with mobile devices such as smartphones. We especially focus on distinction between cars and motorbikes, which have been a difficult problem due to their similarity. In order to solve this problem, we propose the classification method based on the model of vibration generation while vehicles are running. We conducted an experiment to evaluate the classification accuracy and obtained the accuracy of over 80%.
The increase in number of vehicles has created problems in many cities across the globe. Building comprehensive knowledge base about global city dynamics and traffic distribution is a key step to provide fundamental solution to the problems. In this paper, we examine a readily available data source; the existing infrastructure of traffic cameras around the world. We have collected real time traffic data from 2,700 public online traffic camera distributed across 10 cities in four continents for a duration of six months. Our platform allows us to automatically search public cameras, collect and process imagery data, remove outliers, and extract traffic density from those images in a highly scalable way. A time series model employing a co-integrated vector autoregression model is presented in which traffic forecasts may be produced and regions of the city not well observed may be suggested. In addition, a topological comparison of six of these networks is presented.
In this paper, we presented a technical framework to calculate the turn delays on road network with floating car data (FCD). Firstly the original FCD collected with GPS equipped taxies was cleaned and matched to a street map with a distributed system based on Hadoop and MongoDB. Secondly the refined dataset was distributed and matched to the specific intersections among 96 time intervals (from 0:00 to 23:59 per 15 minutes). Thirdly a principal curve-based algorithm was presented to estimate the turn delays at the given intersections. The algorithm is argued not only statistically fitted the real traffic conditions but also is insensitive to data sparseness and data missing problems. We adopted the floating car data collected from March to June in Beijing in 2011, which contains more than 2.6 million trajectories generated from about 20,000 GPS-equipped taxicabs and accounts for about 600 GB in data volume. The result shows the presented algorithm takes precedence of traditional methods, such as mean and median based approaches, and holds a higher estimation accuracy (about 10%~15% higher in RMSE), as well as reflects the changing trend of traffic congestion. With the estimation result, the turn delay ratios both on the whole network and on the 400 main intersections are calculated. It indicates that average 60% of the travel time on the road network, especially in daytime, is cost on intersections, and the 400 main intersections, which only take 2.7% of all the intersections, yet cost about 18% travel time in Beijing.
Smartphones and other vehicular sensors equipped with GPS and wireless networking capabilities, are becoming ubiquitous in transportation systems. They provide us with opportunities to gather timely information about road traffic conditions, fuse (assimilate) it with traffic flow models to improve upon the accuracy of these models, and hence supply valuable information for real-time transportation decision making. Macroscopic traffic flow models are described by systems of partial differential equations (PDEs), which are usually only solved numerically. Adaptive moving mesh methods have shown promise in handling high variability of the spatio-temporal features (e.g. shocks and discontinuities) in model's solutions. We propose a novel low-overhead strategy to adaptively select observation sites in real time, by relying on information from the adaptive moving mesh of the numerical solver of the underlying PDEs. The idea is to place more of the limited observational resources to locations of higher variability in the numerical solution. We incorporate our strategy into a particle-filter based data assimilation framework, and compare it with the strategy of gathering and assimilating measurements from evenly spaced observation sites. We experimentally show that our strategy reduces the relative error by up to 53% in estimating vehicle density on a road during phantom jams and traffic jams due to bottlenecks.
Interpolation of motion data is a challenging problem that is often overlooked by researchers when using GPS data with low sampling rate. In this paper we define a spatio-temporal query called Historical Spatio-Temporal Query (HSTQ). This query receives a time and returns a waypoint containing spatio-temporal data of a moving object at that particular time. To respond to the query for any given time (e.g. every second), we need to interpolate missing waypoints of the trajectory. Linear Interpolation (LI) is the most commonly used method although it can be grossly inaccurate. To deal with this problem, we propose a method called Map-Based Interpolation (MBI). This method uses routing web-services to find significant points (path) between two waypoints. Instead of using road networks, which is typically used by methods such as map-matching, we send a query to a routing web service and analyse the returned data to find missing waypoints. To be able to respond to queries sent to HSTQ for any input time within a trajectory period, we use a combination of MBI and LI (MBI+LI) to interpolate the missing data. We also propose two measures for comparing performance of our interpolation method with LI. Experimental results using realistic trajectories show significant improvement in quality and accuracy of interpolated and down-sampled trajectory data using HSTQ with MBI+LI in comparison to the earlier, widely used LI method.
In this paper we deal with a strategy for a collaborative positioning of vehicles to improve their ego positioning capabilities. One way to achieve this is the sharing of the vehicle's own position and additional measurements to vehicles with known position in their surrounding area. Under the assumption that a single vehicle is able to obtain its ego position by on-board sensors (like laser scanners and GNSS equipment) and in combination with available landmark maps, the consideration of additional measurements to other vehicles leads to a position improvement especially in case of sparse landmark maps. Based on an available landmark map covering built-up areas and highway-like roads, a set of simulations is carried out to evaluate the resulting improvement by using relative position data among nearby vehicles. Different kinds of collaborative positioning scenarios are investigated and contrasted with ego positioning using only the landmark map.
For some graph classes, most notably real-world road networks, shortest path queries can be answered very efficiently if the graph is preprocessed into a contraction hierarchy. The preprocessing algorithm contracts nodes in some order, adding new edges (shortcuts) in the process. While preprocessing and query algorithm work for any contraction ordering, it is desirable to use one that produces as few shortcuts as possible. It is known that the problem of minimizing the size (number of edges) of a given graph's contraction hierarchy is APX-hard. Also, any graph can be processed into a contraction hierarchy with at most O(nh log D) edges, where n, D, and h are the number of nodes, the diameter, and the highway dimension of the original graph, respectively. In this paper we show that the O(nh log D) bound is tight for a wide range of parameters n, D, and h. We also show that planar graphs, despite having highway dimension Ω(√n), can be preprocessed into a graph of size O(n log n). Finally, we present a simpler proof of APX-hardness.
