Search

Sustainable concepts for on-demand transportation, such as ridesharing or ridehailing, require advanced technologies and novel dynamic planning and prediction methods. In this paper, we consider the prediction of taxi trip durations, focusing on the problem of the estimated time of arrival (ETA). ETA can be used to compute and compare alternative taxi schedules and to provide information to drivers and passengers. To solve the underlying hard computational problem with high precision, machine learning (ML) models for ETA are the state of the art. However, these models are mostly black box neural networks. Hence, the resulting predictions are difficult to explain to users. To address this problem, the contributions of this paper are threefold. First, we propose a novel stacked two-level ensemble model combining multiple ETA models; we show that the stacked model outperforms state-of-the-art ML models. However, the complex ensemble architecture makes the resulting predictions less transparent. To alleviate this, we investigate explainable artificial intelligence (XAI) methods for explaining the first- and second-level models of the ensemble. Third, we consider and compare different ways of combining first-level and second-level explanations. This novel concept enables us to explain stacked ensembles for regression tasks. The experimental evaluation indicates that the considered ETA models correctly learn the importance of those input features driving the prediction.

Feeling safe is a major issue for cyclists, and some potential cyclists are still deterred from using the bicycle because they feel too unsafe. Assessing the subjective safety of existing cycling infrastructures and locations can be done by questionnaires that show pictures of infrastructures and ask participants for their safety ratings. However, future cycling infrastructures should also be evaluated as safe even before they are implemented. Therefore, it is desirable to have a method that is able to predict safety from infrastructural information. This study aims to propose two different ways for such a method and to test both ways in a use case. We first developed two scores, namely the Repertory Grid (RG) Score and the FixMyBerlin (FMB) Score, which predict subjective safety from objective environmental information but use different data bases and different methodologies. In a second step, we validated these scores by comparing them to questionnaire ratings that evaluated cyclists’ subjective safety at 20 locations in the city of Braunschweig, Germany. Finally, we compared the two scores as well as the questionnaire ratings with objective safety measures, namely crash statistics, at the respective locations. The results show that the RG Score has a moderate agreement and the FMB Score has a fair agreement with the questionnaire ratings. All methods agree on the overall safety evaluation of various cycling facilities. However, the RG Score showed less variance in the safety ratings, whereas the FMB Score rated most locations more unsafe than the participants in the questionnaire. Interestingly, neither the scores nor the questionnaire ratings could sufficiently deduce the occurrence of a crash at one of the locations. The findings strengthen the importance of subjective safety as a construct independent of objective safety. Furthermore, they provide insights into aspects of subjective safety that can easily be measured by objective scores, and into aspects that are important for cyclists but were not yet covered by the scores. This study, therefore, provides a basis for future considerations and future evaluation methods to assess the subjective safety of cyclists.

Deep learning models need a lot of labeled data to work well. In this study, we use a Self-Supervised Learning (SSL) method for semantic segmentation of archaeological monuments in Digital Terrain Models (DTMs). This method first uses unlabeled data to pretrain a model (pretext task), and then fine-tunes it with a small labeled dataset (downstream task). We use unlabeled DTMs and Relief Visualizations (RVs) to train an encoder-decoder and a Generative Adversarial Network (GAN) in the pretext task and an annotated DTM dataset to fine-tune a semantic segmentation model in the downstream task. Experiments indicate that this approach produces better results than training from scratch or using models pretrained on image data like ImageNet. The code and pretrained weights for the encoder-decoder and the GAN models are made available on Github.1

Shared mobility has emerged as a sustainable alternative to both private transportation and traditional public transport, promising to reduce the number of private vehicles on roads while offering users greater flexibility. Today, urban areas are home to a myriad of innovative services, including car-sharing, ride-sharing, and micromobility solutions like moped-sharing, bike-sharing, and e-scooter-sharing. Given the intense competition and the inherent operational complexities of shared mobility systems, providers are increasingly seeking specialized decision-support methodologies to boost operational efficiency. While recent research indicates that advanced machine learning methods can tackle the intricate challenges in shared mobility management decisions, a thorough evaluation of existing research is essential to fully grasp its potential and pinpoint areas needing further exploration. This paper presents a systematic literature review that specifically targets the application of Machine Learning for decision-making in Shared Mobility Systems. Our review underscores that Machine Learning offers methodological solutions to specific management challenges crucial for the effective operation of Shared Mobility Systems. We delve into the methods and datasets employed, spotlight research trends, and pinpoint research gaps. Our findings culminate in a comprehensive framework of Machine Learning techniques designed to bolster managerial decision-making in addressing challenges specific to Shared Mobility across various levels.

As urbanisation increases, in many places, the transport system is suffering from problems that may affect large parts of the urban population, such as traffic congestion or increased air pollution. In both cases, a better distribution of traffic flows could contribute to establishing a more sustainable transport system, and to improve the situation from a societal point of view. In this paper, we use cartographic symbolisation for communicating favourability of route options for achieving a societal benefit. Since map symbols can evoke different emotional responses in the viewer, we investigate to which extent map symbols evoke positive and negative emotions and whether these influence route choice decision making. We created different cartographic visualisations and designed a user study that investigates the effectiveness and suitability of these different visualisation variants for influencing route choice based on two scenarios: traffic and air quality. Fourteen route maps were prepared using different map symbols to symbolise societally favourable and non-favourable route options. The results of this study show that map symbols can be used effectively for influencing route choice towards choosing the favourable route for the two tested scenarios. While visualisations that modify only lines were more effective in the traffic scenario, area symbol modifications were more effective for the air quality scenario. The symbolisation evoked a wide range of emotions in participants. While non-favourable routes mainly evoke negative emotions (particularly fear), favourable routes mainly evoked positive emotions (particularly contentment) or no emotions. The results further demonstrate that for some of the visualisation variants, emotions felt in response to the map visualisations contributed significantly to changing the route choice decisions in favour of the societally favourable route option. The findings of this research demonstrate the relationship between route choice behaviour and emotional responses elicited by map symbols.

In this paper, we focus on trajectories at intersections regulated by various regulation types such as traffic lights, priority/yield signs, and right-of-way rules. We test some methods to detect and recognize movement patterns from GPS trajectories, in terms of their geometrical and spatio-temporal components. In particular, we first find out the main paths that vehicles follow at such locations. We then investigate the way that vehicles follow these geometric paths (how do they move along them). For these scopes, machine learning methods are used and the performance of some known methods for trajectory similarity measurement (DTW, Hausdorff, and Fréchet distance) and clustering (Affinity propagation and Agglomerative clustering) are compared based on clustering accuracy. Afterward, the movement behavior observed at six different intersections is analyzed by identifying certain movement patterns in the speed- and time-profiles of trajectories. We show that depending on the regulation type, different movement patterns are observed at intersections. This finding can be useful for intersection categorization according to traffic regulations. The practicality of automatically identifying traffic rules from GPS tracks is the enrichment of modern maps with additional navigation-related information (traffic signs, traffic lights, etc.).

This short paper gives a subjective view on cartographic generalization, its achievements in the past, and the challenges it faces in the future.

An example of automated characterization and interpretation of the textural and compositional characteristics of solids phases in thin sections using machine learning (ML) is presented. In our study, we focus on the characterization of olivine in volcanic rocks, which is a phase that is often chemically zoned with variable Mg/(Mg + Fe) ratios, so-called magnesian number or mg#. As the olivine crystals represent only less than 10 vol% of the volcanic rock, a pre-processing step is necessary to automatically detect the phases of interest in the images on a pixel level, which is achieved using Deep Learning. A major contribution of the presented approach is to use backscattered electron (BSE) images to: 1) automatically segment all olivine crystals present in the thin section; 2) determine quantitatively their mg#; and 3) identify different populations depending on zoning type (e.g., normal vs reversal zoning) and textural characteristics (e.g., microlites vs phenocrysts). The segmentation of the olivine crystals is implemented with a pretrained fully convolutional neural network model with DeepLabV3 architecture. The model is trained to identify olivine crystals in backscattered electron images using automatically generated training data. The training data are generated automatically from images which can easily be created from X-Ray element maps. Once the olivines are identified in the BSE images, the relationship between BSE intensity value and mg# is determined using a simple regression based on a set of microprobe measurements. This learned functional relationship can then be applied to all olivine pixels of the thin section. If the highest possible map resolution (1 micron per 1 pixel) is selected for the data acquisition, the full processing time of an entire thin section of ∼3×4cm containing more than 1,500 phenocrysts and 20.000 microliths required 140 h of data acquisition (BSE + X-Ray element maps), 8 h of training and 16 h of segmentation and classification. Our further tests demonstrated that the 140 h of data acquisition can be reduced at least by a factor of 4 since only a part of the thin section area (25% or even less) needs to be used for training. The characterization of each additional thin section would only require the BSE data acquisition time (less than 48 h for a whole thin section), without an additional training step. The paper describes the training and processing in detail, shows analytical results and outlines the potential of this Deep Learning approach for petrological applications, resulting in the automatic characterization and interpretation of mineral textures and compositions with an unprecedented high resolution.

Nowadays mobile positioning devices, such as global navigation satellite systems (GNSS) but also external sensor technology like cameras allow an efficient online collection of trajectories, which reflect the behavior of moving objects, such as cars. The data can be used for various applications, e.g., traffic planning or updating maps, which need many trajectories to extract and infer the desired information, especially when machine or deep learning approaches are used. Often, the amount and diversity of necessary data exceeds what can be collected by individuals or even single companies. Currently, data owners, e.g., vehicle producers or service operators, are reluctant to share data due to data privacy rules or because of the risk of sharing information with competitors, which could jeopardize the data owner's competitive advantage. A promising approach to exploit data from several data owners, but still not directly accessing the data, is the concept of federated learning, that allows collaborative learning without exchanging raw data, but only model parameters.In this paper, we address the problem of anomaly detection in vehicle trajectories, and investigate the benefits of using federated learning. To this end, we apply several state-of-the-art learning algorithms like one-class support vector machine (OCSVM) and isolation forest, thus solving a one-class classification problem. Based on these learning mechanisms, we successfully proposed and verified a federated architecture for the collaborative identification of anomalous trajectories at several intersections. We demonstrate that the federated approach is beneficial not only to improve the overall anomaly detection accuracy, but also for each individual data owner. The experiments show that federated learning allows to increase the anomaly detection accuracy from in average AUC-ROC scores of 97% by individual intersections up to 99% using cooperation.

In this paper, a new method is proposed to detect traffic regulations at intersections using GPS traces. The knowledge of traffic rules for regulated locations can help various location-based applications in the context of Smart Cities, such as the accurate estimation of travel time and fuel consumption from a starting point to a destination. Traffic regulations as map features, however, are surprisingly still largely absent from maps, although they do affect traffic flow which, in turn, affects vehicle idling time at intersections, fuel consumption, CO2 emissions, and arrival time. In addition, mapping them using surveying equipment is costly and any update process has severe time constraints. This fact is precisely the motivation for this study. Therefore, its objective is to propose an automatic, fast, scalable, and inexpensive way to identify the type of intersection control (e.g., traffic lights, stop signs). A new method based on summarizing the collective behavior of vehicle crossing intersections is proposed. A modification of a well-known clustering algorithm is used to detect stopping and deceleration episodes. These episodes are then used to categorize vehicle crossing of intersections into four possible traffic categories (p1: free flow, p2: deceleration without stopping events, p3: only one stopping event, p4: more than one stopping event). The percentages of crossings of each class per intersection arm, together with other speed/stop/deceleration features, extracted from trajectories, are then used as features to classify the intersection arms according to their traffic control type (dynamic model). The classification results of the dynamic model are compared with those of the static model, where the classification features are extracted from OpenStreetMap. Finally, a hybrid model is also tested, where a combination of dynamic and static features is used, which outperforms the other two models. For each of the three models, two variants of the feature vector are tested: one where only features associated with a single intersection arm are used (one-arm model) and another where features also from neighboring intersection arms of the same intersection are used to classify an arm (all-arm model). The methodology was tested on three datasets and the results show that all-arm models perform better than single-arm models with an accuracy of 95% to 97%.

The increasing availability of devices to capture the position of moving objects (and other environmental information) leads to a very large amount and variety of mobility data. In order to obtain important information about the objects, their behavior or the environment of the objects, an automatic analysis is required. This article highlights current research questions in the context of the analysis of mobility data and presents them on the basis of work carried out at the Institute of Cartography and Geoinformatics (ikg) at Leibniz University of Hannover, Germany. A focus is put on the analysis and exploitation of information from Mobile Mapping vehicles.

Accurate information of traffic regulators at junctions is important for navigating and driving in cities. However, such information is often missing, incomplete or not up-to-date in digital maps due to the high cost, e.g., time and money, for data acquisition and updating. In this study we propose a crowdsourced method that harnesses the light-weight GPS tracks from commuting vehicles as Volunteered Geographic Information (VGI) for traffic regulator detection. We explore the novel idea of detecting traffic regulators by learning the movement patterns of vehicles at regulated locations. Vehicles’ movement behavior was encoded in the form of speed-profiles, where both speed values and their sequential order during movement development were used as features in a three-class classification problem for the most common traffic regulators: traffic-lights, priority-signs and uncontrolled junctions. The method provides an average weighting function and a majority voting scheme to tolerate the errors in the VGI data. The sequence-to-sequence framework requires no extra overhead for data processing, which makes the method applicable for real-world traffic regulator detection tasks. The results showed that the deep-learning classifier Conditional Variational Autoencoder can predict regulators with 90% accuracy, outperforming a random forest classifier (88% accuracy) that uses the summarized statistics of movement as features. In our future work images and augmentation techniques can be leveraged to generalize the method’s ability for classifying a greater variety of traffic regulator classes.

Tracking moving objects is both challenging and important for a large variety of applications. Different technologies based on the global positioning system (GPS) and video or radio data are used to obtain the trajectories of the observed objects. However, in some use cases, they fail to provide sufficiently accurate, complete and correct data at the same time. In this work we present an approach for fusing GPS- and video-based tracking in order to exploit their individual advantages. In this way we aim to combine the reliability of GPS tracking with the high geometric accuracy of camera detection. For the fusion of the movement data provided by the different devices we use a hidden Markov model (HMM) formulation and the Viterbi algorithm to extract the most probable trajectories. In three experiments, we show that our approach is able to deal with challenging situations like occlusions or objects which are temporarily outside the monitored area. The results show the desired increase in terms of accuracy, completeness and correctness.

Mapping with surveying equipment is a time-consuming and cost-intensive procedure that makes the frequent map updating unaffordable. In the last few years, much research has focused on eliminating such problems by counting on crowdsourced data, such as GPS traces. An important source of information in maps, especially under the consideration of forthcoming self-driving vehicles, is the traffic regulators. This information is largely lacking in maps like OpenstreetMap (OSM) and this article is motivated by this fact. The topic of this systematic literature review (SLR) is the detection and recognition of traffic regulators such as traffic lights (signals), stop-, yield-, priority-signs, right of way priority rules and turning restrictions at intersections, by leveraging non imagery crowdsourced data. More particularly, the aim of this study is (1) to identify the range of detected and recognised regulatory types by crowdsensing means, (2) to indicate the different classification techniques that can be used for these two tasks, (3) to assess the performance of different methods, as well as (4) to identify important aspects of the applicability of these methods. The two largest databases of peer-reviewed literature were used to locate relevant research studies and after different screening steps eleven articles were selected for review. Two major findings were concluded—(a) most regulator types can be identified with over 80% accuracy, even using heuristic-driven approaches and (b) under the current progress on the field, no study can be reproduced for comparative purposes nor can solely rely on open data sources due to lack of publicly available datasets and ground truth maps. Future research directions are highlighted as possible extensions of the reviewed studies.

Cartographic generalization is a problem, which poses interesting challenges to automation. Whereas plenty of algorithms have been developed for the different sub-problems of generalization (e.g., simplification, displacement, aggregation), there are still cases, which are not generalized adequately or in a satisfactory way. The main problem is the interplay between different operators. In those cases the human operator is the benchmark, who is able to design an aesthetic and correct representation of the physical reality. Deep learning methods have shown tremendous success for interpretation problems for which algorithmic methods have deficits. A prominent example is the classification and interpretation of images, where deep learning approaches outperform traditional computer vision methods. In both domains-computer vision and cartography-humans are able to produce good solutions. A prerequisite for the application of deep learning is the availability of many representative training examples for the situation to be learned. As this is given in cartography (there are many existing map series), the idea in this paper is to employ deep convolutional neural networks (DCNNs) for cartographic generalizations tasks, especially for the task of building generalization. Three network architectures, namely U-net, residual U-net and generative adversarial network (GAN), are evaluated both quantitatively and qualitatively in this paper. They are compared based on their performance on this task at target map scales 1:10,000, 1:15,000 and 1:25,000, respectively. The results indicate that deep learning models can successfully learn cartographic generalization operations in one single model in an implicit way. The residual U-net outperforms the others and achieved the best generalization performance.

This report of a community activity, a Dagstuhl Seminar earlier in 2010, postulates the need for a computational transportation science, as the science behind intelligent transportation systems. In addition to the argument for establishing a discipline, we present a first research agenda for computational transportation science.

Global Navigation Satellite Systems (GNSS) deliver absolute position and velocity, as well as time information (P, V, T). However, in urban areas, the GNSS navigation performance is restricted due to signal obstructions and multipath. This is especially true for applications dealing with highly automatic or even autonomous driving. Subsequently, multi-sensor platforms including laser scanners and cameras, as well as map data are used to enhance the navigation performance, namely in accuracy, integrity, continuity and availability. Although well-established procedures for integrity monitoring exist for aircraft navigation, for sensors and fusion algorithms used in automotive navigation, these concepts are still lacking. The research training group i.c.sens, integrity and collaboration in dynamic sensor networks, aims to fill this gap and to contribute to relevant topics. This includes the definition of alternative integrity concepts for space and time based on set theory and interval mathematics, establishing new types of maps that report on the trustworthiness of the represented information, as well as taking advantage of collaboration by improved filters incorporating person and object tracking. In this paper, we describe our approach and summarize the preliminary results.

In recent years, pluvial floods caused by extreme rainfall events have occurred frequently. Especially in urban areas, they lead to serious damages and endanger the citizens’ safety. Therefore, real-time information about such events is desirable. With the increasing popularity of social media platforms, such as Twitter or Instagram, information provided by voluntary users becomes a valuable source for emergency response. Many applications have been built for disaster detection and flood mapping using crowdsourcing. Most of the applications so far have merely used keyword filtering or classical language processing methods to identify disaster relevant documents based on user generated texts. As the reliability of social media information is often under criticism, the precision of information retrieval plays a significant role for further analyses. Thus, in this paper, high quality eyewitnesses of rainfall and flooding events are retrieved from social media by applying deep learning approaches on user generated texts and photos. Subsequently, events are detected through spatiotemporal clustering and visualized together with these high quality eyewitnesses in a web map application. Analyses and case studies are conducted during flooding events in Paris, London and Berlin.

Physical environmental processes, such as the evolution of precipitation or the diffusion of chemical clouds in the atmosphere, can be approximated by numerical models based on the underlying physics, e.g., for the purpose of prediction. As the modeling process is often very complex and resource demanding, such models are sometimes replaced by those that use historic and current data for calibration. For atmospheric (e.g., precipitation) or oceanographic (e.g., sea surface temperature) fields, the data-driven methods often concern the horizontal displacement driven by transport processes (called advection). These methods rely on flow fields estimated from images of the phenomenon by computer vision techniques, such as optical flow (OF). In this work, an algorithm is proposed for estimating the motion of spatio-temporal fields with the nodes of a geosensor network (GSN) deployed in situ when images are not available. The approach adapts a well-known raster-based OF algorithm to the specifics of GSNs, especially to the spatial irregularity of data. In this paper, the previously introduced approach has been further developed by introducing an error model that derives probabilistic error measures based on spatial node configuration. Further, a more generic motion model is provided, as well as comprehensive simulations that illustrate the performance of the algorithm in different conditions (fields, motion behaviors, node densities and deployments) for the two error measures of motion direction and motion speed. Finally, the algorithm is applied to data sampled from weather radar images, and the algorithm performance is compared to that of a state-of-the-art OF algorithm applied to the weather radar images directly, as often done in nowcasting.