Cooperative perception can increase the view field and decrease the occlusion of an ego vehicle, hence improving the perception performance and safety of autonomous driving. Despite the success of previous works on cooperative object detection, they mostly operate on dense Bird’s Eye View (BEV) feature maps, which are computationally demanding and can hardly be extended to long-range detection problems. More efficient fully sparse frameworks are rarely explored. In this work, we design a fully sparse framework, SparseAlign, with three key features: an enhanced sparse 3D backbone, a query-based temporal context learning module, and a robust detection head specially tailored for sparse features. Extensive experimental results on both OPV2V and DAIR-V2X datasets show that our framework, despite its sparsity, outperforms the state of the art with less communication bandwidth requirements. In addition, experiments on the OPV2Vt and DAIR-V2Xt datasets for time-aligned cooperative object detection also show a significant performance gain compared to the baseline works.

Accurate surface normal estimation is critical for LiDAR-based tasks such as mapping, localization, and scene understanding. However, traditional 3D neighborhood selection methods often struggle with sparse point clouds due to uneven density and sensor limitations. In this paper, we propose a geometry-aware approach that leverages the structured 2D range view to select more consistent neighborhoods for normal estimation. By combining adaptive windowing in range space with distance-based filtering, our method provides robust and accurate normal vectors across both dense and sparse regions. We integrate this method into a lightweight SLAM framework and evaluate it using indoor LiDAR data from a Velodyne-16 sensor. The results demonstrate improved normal estimation and enhanced localization and mapping performance, highlighting the method’s effectiveness for indoor robotic applications.

In autonomous driving tasks, trajectory prediction in complex traffic environments requires adherence to real-world context conditions and behavior multimodalities. Existing methods predominantly rely on prior assumptions or generative models trained on curated data to learn road agents’ stochastic behavior bounded by scene constraints. However, they often face mode averaging issues due to data imbalance and simplistic priors, and could even suffer from mode collapse due to unstable training and single ground truth supervision. These issues lead the existing methods to a loss of predictive diversity and adherence to the scene constraints. To address these challenges, we introduce a novel trajectory generator named Controllable Diffusion Trajectory (CDT), which integrates map information and social interactions into a Transformer-based conditional denoising diffusion model to guide the prediction of future trajectories. To ensure multimodality, we incorporate behavioral tokens to direct the trajectory’s modes, such as going straight, turning right or left. Moreover, we incorporate the predicted endpoints as an alternative behavioral token into the CDT model to facilitate the prediction of accurate trajectories. Extensive experiments on the Argoverse 2 benchmark demonstrate that CDT excels in generating diverse and scene-compliant trajectories in complex urban settings.

Collective Perception has attracted significant attention in recent years due to its advantage for mitigating occlusion and expanding the field-of-view, thereby enhancing reliability, efficiency, and, most crucially, decision-making safety. However, developing collective perception models is highly resource demanding due to extensive requirements of processing input data for many agents, usually dozens of images and point clouds for a single frame. This not only slows down the model development process for collective perception but also impedes the utilization of larger models. In this paper, we propose an agent-based training framework that handles the deep learning modules and agent data separately to have a cleaner data flow structure. This framework not only provides an API for flexibly prototyping the data processing pipeline and defining the gradient calculation for each agent, but also provides the user interface for interactive training, testing and data visualization. Training experiment results of four collective object detection models on the prominent collective perception benchmark OPV2V show that the agent-based training can significantly reduce the GPU memory consumption and training time while retaining inference performance. The framework and model implementations are available at https://github.com/YuanYunshuang/CoSense3D

Predicting trajectories of pedestrians based on goal information in highly interactive scenes is a crucial step toward Intelligent Transportation Systems and Autonomous Driving. The challenges of this task come from two key sources: (1) complex social interactions in high pedestrian density scenarios and (2) limited utilization of goal information to effectively associate with past motion information. To address these difficulties, we integrate social forces into a Transformer-based stochastic generative model backbone and propose a new goal-based trajectory predictor called ForceFormer. Differentiating from most prior works that simply use the destination position as an input feature, we leverage the driving force from the destination to efficiently simulate the guidance of a target on a pedestrian. Additionally, repulsive forces are used as another input feature to describe the avoidance action among neighboring pedestrians. Extensive experiments show that our proposed method achieves on-par performance measured by distance errors with the state-of-the-art models but evidently decreases collisions, especially in dense pedestrian scenarios on widely used pedestrian datasets.

The shared space design is applied in urban streets to support barrier-free movement and integrate traffic participants (such as pedestrians, cyclists and vehicles) into a common road space. Regardless of the low-speed environment, sharing space with motor vehicles can make vulnerable road users feel uneasy. Yet, walking in groups increases their confidence as well as influence the yielding behavior of drivers. Therefore, we propose an innovative approach to support the crossing of pedestrians via grouping and project the virtual lanes in shared spaces. This paper presents the important components of the crowd steering system, discusses the enablers and gaps in the current approach, and illustrates the proposed idea with concept diagrams.

The idea of ‘citizen as sensors’ has gradually become a reality over the past decade. Today, Volunteered Geographic Information (VGI) from citizens is highly involved in acquiring information on natural disasters. In particular, the rapid development of deep learning techniques in computer vision and natural language processing in recent years has allowed more information related to natural disasters to be extracted from social media, such as the severity of building damage and flood water levels. Meanwhile, many recent studies have integrated information extracted from social media with that from other sources, such as remote sensing and sensor networks, to provide comprehensive and detailed information on natural disasters. Therefore, it is of great significance to review the existing work, given the rapid development of this field. In this review, we summarized eight common tasks and their solutions in social media content analysis for natural disasters. We also grouped and analyzed studies that make further use of this extracted information, either standalone or in combination with other sources. Based on the review, we identified and discussed challenges and opportunities.

To accurately predict future positions of different agents in traffic scenarios is crucial for safely deploying intelligent autonomous systems in the real-world environment. However, it remains a challenge due to the behavior of a target agent being affected by other agents dynamically and there being more than one socially possible paths the agent could take. In this paper, we propose a novel framework, named Dynamic Context Encoder Network (DCENet). In our framework, first, the spatial context between agents is explored by using self-attention architectures. Then, the two-stream encoders are trained to learn temporal context between steps by taking the respective observed trajectories and the extracted dynamic spatial context as input. The spatial-temporal context is encoded into a latent space using a Conditional Variational Auto-Encoder (CVAE) module. Finally, a set of future trajectories for each agent is predicted conditioned on the learned spatial-temporal context by sampling from the latent space, repeatedly. DCENet is evaluated on one of the most popular challenging benchmarks for trajectory forecasting Trajnet and reports a new state-of-the-art performance. It also demonstrates superior performance evaluated on the benchmark inD for mixed traffic at intersections. A series of ablation studies is conducted to validate the effectiveness of each proposed module. Our code is available at https://github.com/wtliao/DCENet.

In shared space, pedestrians are often found walking in groups and behaving differently than individual pedestrians. However, automatically detecting pedestrian groups with high accuracy is not trivial given the dynamic environment and interactions in mixed traffic. Instead of tedious manual work and in order to cope with large scales of data, we propose a time-sequence Density-Based Spatial Clustering of Applications with Noise (DBSCAN) for pedestrian group detection. It is based on coexisting time and Euclidean distance between pedestrians. Our approach outputs reliable results with high IoU values. It can be easily adapted to other groups, e.g., cyclists and animals. In addition to individual behavior, the output data with differentiation of group behavior can be used in further studies in intent detection and motion prediction.

Since in many cities transport infrastructure is operating at or beyond capacity, novel approaches to organize urban mobility are gaining attraction. However, assessing the benefits of a measure that has disruptive capacity in a complex system requires a carefully designed research. This paper takes a recent idea for urban mobility - flexible road trains - and illustrates the computational and research challenges of realizing its full potential and describing its social, ecological and economical impact.

This work aims at the inference of traffic networks from GPS trajectories. We perform geometry and topology reconstruction of the network in a multistep process. Our main contributions are the formulation of an explicit intersection model with a score function that accounts for consistency with the raw tracking data, as well as for a topology prior and the search for the best model by maximization of this score function using a Markov chain Monte Carlo sampler. We demonstrate the viability of our model-based approach with experiments on GPS data sets of varying size and data quality, followed by a comparison with results achieved by alternative, heuristic approaches.

Accurate estimations of spatio-temporal fields at unsampled locations are important in a number of applications. Often, spatio-temporal fields are advected, which means the change in field values over time at a particular point in space stems to a large extent from motion of a more or less constant spatial field. For such dynamic fields, interpolation methods including information on the motion behaviour of the field are promising extensions of solely spatial (snapshot) and symmetric spatio-temporal methods. In this paper, the performance of different deterministic and geostatistical interpolation methods is compared for precipitation estimation from 1-minute time series of spatially distributed rain gauges. The focus is on spatio-temporal methods that include information on the motion behaviour of the rainfield, estimated from weather radar using optical flow. The different interpolation methods are introduced and evaluated using rain gauge measurements of a 15-day period and cross-validation. The results show that including information on the motion behaviour significantly improves interpolation quality in terms of RMSE.

In this paper we aim to recognize a priori unknown group movement patterns. We propose a constellation-based approach to extract repetitive relative movements of a constant group, which are allowed to be rotated, translated or differently scaled. To this end, we record a sequence of constellations, which are used for describing the movements relatively. We deal with uncertainties, and similarities of constellations respectively, by clustering the constellations. Further, we have developed a sequence mining algorithm, which uses the clustering results and tree-like data structures to extract the requested patterns from the sequence. Finally, this approach is applied to different datasets containing real trajectory data provided by different tracking devices. By this way, we want to show its portability to different use cases.

This paper presents a generative statistical approach to 3D building roof reconstruction from airborne laser scanning point clouds. In previous works bottom-up methods, e.g., points clustering, plane detection, and contour extraction, are widely used. Since the laser scanning data of urban scenes often contain extra structures and artefacts due to tree clutter, reflection from windows, water features, etc., bottom-up reconstructions may result in a number of incomplete or irregular roof parts.

In recent times the amount of spatial data being collected by voluntary users, e.g. as part of the OpenStreetMap project, is rapidly increasing. Due to the fact, that everyone can participate in this social collaboration, the completeness and accuracy of the data is very heterogeneous. Although a object catalogue exists as part of the OSM project, users are not restricted which attributes they set and to which detail. Therefore the geometry of a feature is more reliable than its attributes. However, in order to use the data for analysis purposes, knowledge about the semantic contents is of importance.

The availability of methods for abstracting and generalizing spatial data is vital for understanding and communicating spatial information. Spatial analysis using maps at different scales is a good example of this. Such methods are needed not only for analogue spatial data sets but even more so for digital data. In order to automate the process of generating different levels of detail of a spatial data set, generalization operations are used. The paper first gives an overview on current approaches for the automation of generalization and data abstraction, and then presents solutions for three generalization problems based on optimization techniques. Least‐Squares Adjustment is used for displacement and shape simplification (here, building groundplans), and Self‐Organizing Maps, a Neural Network technique, is applied for typification, i.e. a density preserving reduction of objects. The methods are validated with several examples and evaluated according to their advantages and disadvantages. Finally, a scenario describes how these methods can be combined to automatically yield a satisfying result for integrating two data sets of different scales.

No abstract available.

In order to handle the ever increasing amount of digital data efficiently, methods and techniques for automation have to be available. A key issue is the use of knowledge-based systems to interpret the data and operate on it in an automatic way. Interpretation requires knowledge about the underlying concepts in the data. Providing the necessary knowledge, however, is a bottleneck in automation. In this paper, an approach to semi-automatic knowledge acquisition is described. The knowledge is derived from given example data using supervised machine learning. An application of the approach in the domain of the interpretation of unstructured spatial data is given, and possible further applications are outlined.