| 1. |
- Kristan, M., et al.
(författare)
-
The Eighth Visual Object Tracking VOT2020 Challenge Results
- 2020
-
Ingår i: Computer Vision. - Cham : Springer International Publishing. - 9783030682378 ; , s. 547-601
-
Konferensbidrag (refereegranskat)abstract
- The Visual Object Tracking challenge VOT2020 is the eighth annual tracker benchmarking activity organized by the VOT initiative. Results of 58 trackers are presented; many are state-of-the-art trackers published at major computer vision conferences or in journals in the recent years. The VOT2020 challenge was composed of five sub-challenges focusing on different tracking domains: (i) VOT-ST2020 challenge focused on short-term tracking in RGB, (ii) VOT-RT2020 challenge focused on “real-time” short-term tracking in RGB, (iii) VOT-LT2020 focused on long-term tracking namely coping with target disappearance and reappearance, (iv) VOT-RGBT2020 challenge focused on short-term tracking in RGB and thermal imagery and (v) VOT-RGBD2020 challenge focused on long-term tracking in RGB and depth imagery. Only the VOT-ST2020 datasets were refreshed. A significant novelty is introduction of a new VOT short-term tracking evaluation methodology, and introduction of segmentation ground truth in the VOT-ST2020 challenge – bounding boxes will no longer be used in the VOT-ST challenges. A new VOT Python toolkit that implements all these novelites was introduced. Performance of the tested trackers typically by far exceeds standard baselines. The source code for most of the trackers is publicly available from the VOT page. The dataset, the evaluation kit and the results are publicly available at the challenge website (http://votchallenge.net ).
|
|
| 2. |
- Yang, Ying, et al.
(författare)
-
Harnessing the power of Machine learning for AIS Data-Driven maritime Research: A comprehensive review
- 2024
-
Ingår i: Transportation Research Part E: Logistics and Transportation Review. - 1366-5545. ; 183
-
Tidskriftsartikel (refereegranskat)abstract
- Automatic Identification System (AIS) data holds immense research value in the maritime industry because of its massive scale and the ability to reveal the spatial–temporal variation patterns of vessels. Unfortunately, its potential has long been limited by traditional methodologies. The emergence of machine learning (ML) offers a promising avenue to unlock the full potential of AIS data. In recent years, there has been a growing interest among researchers in leveraging ML to analyze and utilize AIS data. This paper, therefore, provides a comprehensive review of ML applications using AIS data and offers valuable suggestions for future research, such as constructing benchmark AIS datasets, exploring more deep learning (DL) and deep reinforcement learning (DRL) applications on AIS-based studies, and developing large-scale ML models trained by AIS data.
|
|
| 3. |
- Fang, Shan, et al.
(författare)
-
A Dynamic Transformation Car-Following Model for the Prediction of the Traffic Flow Oscillation
- 2024
-
Ingår i: IEEE Intelligent Transportation Systems Magazine. - 1939-1390 .- 1941-1197. ; 16:1, s. 174-198
-
Tidskriftsartikel (refereegranskat)abstract
- Car-following (CF) behavior is a fundamental of traffic flow modeling; it can be used for the virtual testing of connected and automated vehicles and the simulation of various types of traffic flow, such as free flow and traffic oscillation. Although existing CF models can replicate the free flow well, they are incapable of simulating complicated traffic oscillation, and it is difficult to strike a balance between accuracy and efficiency. This article investigates the error variation when the traffic oscillation is simulated by the intelligent driver model (IDM). Then, it divides the traffic oscillation into four phases (coasting, deceleration, acceleration, and stationary) by using the space headway of multiple steps. To simulate traffic oscillation between multiple human-driven vehicles, a dynamic transformation CF model is proposed, which includes the long-time prediction submodel [modified sequence-to-sequence (Seq2seq)] model, short-time prediction submodel (Transformer), and their dynamic transformation strategy]. The first submodel is utilized to simulate the coasting and stationary phases, while the second submodel is utilized to simulate the acceleration and deceleration phases. The results of experiments indicated that compared to K-nearest neighbors, IDM, and Seq2seq CF models, the dynamic transformation CF model reduces the trajectory error by 60.79–66.69% in microscopic traffic flow simulations, 7.71–29.91% in mesoscopic traffic flow simulations, and 1.59–18.26% in macroscopic traffic flow simulations. Moreover, the runtime of the dynamic transformation CF model (Inference) decreased by 14.43–66.17% when simulating the large-scale traffic flow.
|
|
| 4. |
- Fedriani, Rubén, 1991, et al.
(författare)
-
The SOFIA Massive (SOMA) Star Formation Survey. IV. Isolated Protostars
- 2023
-
Ingår i: Astrophysical Journal. - : American Astronomical Society. - 1538-4357 .- 0004-637X. ; 942:1
-
Tidskriftsartikel (refereegranskat)abstract
- We present similar to 10-40 mu m SOFIA-FORCAST images of 11 isolated protostars as part of the SOFIA Massive (SOMA) Star Formation Survey, with this morphological classification based on 37 mu m imaging. We develop an automated method to define source aperture size using the gradient of its background-subtracted enclosed flux and apply this to build spectral energy distributions (SEDs). We fit the SEDs with radiative transfer models, developed within the framework of turbulent core accretion (TCA) theory, to estimate key protostellar properties. Here, we release the sedcreator python package that carries out these methods. The SEDs are generally well fitted by the TCA models, from which we infer initial core masses M ( c ) ranging from 20-430 M (circle dot), clump mass surface densities sigma(cl) similar to 0.3-1.7 g cm(-2), and current protostellar masses m (*) similar to 3-50 M (circle dot). From a uniform analysis of the 40 sources in the full SOMA survey to date, we find that massive protostars form across a wide range of clump mass surface density environments, placing constraints on theories that predict a minimum threshold sigma(cl) for massive star formation. However, the upper end of the m (*)-sigma(cl) distribution follows trends predicted by models of internal protostellar feedback that find greater star formation efficiency in higher sigma(cl) conditions. We also investigate protostellar far-IR variability by comparison with IRAS data, finding no significant variation over an similar to 40 yr baseline.
|
|
| 5. |
- Fei, Yang, et al.
(författare)
-
Critical Roles of Control Engineering in the Development of Intelligent and Connected Vehicles
- 2024
-
Ingår i: Journal of Intelligent and Connected Vehicles. - 2399-9802. ; 7:2, s. 79-85
-
Forskningsöversikt (refereegranskat)abstract
- In recent years, advancements in onboard computing hardware and wireless communication technology have remarkably stimulated the development of intelligent and connected vehicles (ICVs). Specifically, some researchers have investigated the issue of employing various advanced control techniques to optimize the performance of autonomous vehicles in practice (Sun et al., 2023; Zhang et al., 2023a, 2023b). Therefore, this article aims to discuss why and how control engineering plays an essential role in the development of ICVs.
|
|
| 6. |
- Figalova, Nikol, et al.
(författare)
-
An extension of the human-factors methodological toolbox for human-AV interaction design research : Deliverable 1.4 in the EC ITN project SHAPE-IT
- 2023
-
Rapport (övrigt vetenskapligt/konstnärligt)abstract
- Early state researchers (ESRs) of the SHAPE-IT project have committed to exploring innovative methods to ensure driving safety during interactions between human and Automated Vehicles (AVs). In this deliverable, insights of ESRs span a broad spectrum of methodologies, from experimental methods, including psychophysiological measures, Virtual Reality/Augmented Reality (VR/AR) applications, and transparency assessments, to human-AV interaction models, with vehicle-pedestrian model and vehicle-cyclist model, and lastly the long-term effects. New types of interactions between humans and AVs need to be evaluated during the systems’ development to ensure that requirements of safety, acceptance, and efficiency are met before they are introduced to the market. Since innovative concepts require great cost and effort for their realization, it is necessary to ascertain whether the expected effects will be achieved. Many of the systems’ ergonomic requirements can be considered using experimental methods based on theoretical knowledge. This proposal outlines different aspects for empirical investigations related to the interaction between human and AV. It is important to mention that different human roles need to be considered inside (passenger or driver) and outside/around (VRU) the AV. The research aspects range from cognitive processes (perception and decision), via motion behavior, to learning and behavioral adaptation. This requires that dedicated methods with clear, consistent definitions be refined or developed. One example is the usage of virtual reality to investigate the complex interaction processes between AVs and VRUs in a safe and controllable setting as an alternative to field trials. Also, different AV communication strategies can be implemented in VR quicker and with reduced effort compared to hardware setups or experimental cars. Further methods are physiological measurements, different types of driving simulation and long-term behavioral study approaches. In their combination the different methods represent a toolbox of methodological approaches to analyze and evaluate different aspects of automated driving realizations. This deliverable presents a collection of recommended experimental approaches that address complex questions using advanced measurement equipment and statistical approaches, and their successful application within the SHAPE-IT project.
|
|
| 7. |
- Figalova, Nikol, et al.
(författare)
-
Methodological Framework for Modelling and Empirical Approaches (Deliverable D1.1 in the H2020 MSCA ITN project SHAPE-IT)
- 2021
-
Rapport (övrigt vetenskapligt/konstnärligt)abstract
- The progress in technology development over the past decades, both with respect to software and hardware, offers the vision of automated vehicles as means of achieving zero fatalities in traffic. However, the promises of this new technology – an increase in road safety, traffic efficiency, and user comfort – can only be realized if this technology is smoothly introduced into the existing traffic system with all its complexities, constraints, and requirements. SHAPE- IT will contribute to this major undertaking by addressing research questions relevant for the development and introduction of automated vehicles in urban traffic scenarios. Previous research has pointed out several research areas that need more attention for a successful implementation and deployment of human-centred vehicle automation in urban environments. In SHAPE-IT, for example, a better understanding of human behaviour and the underlying psychological mechanisms will lead to improved models of human behaviour that can help to predict the effects of automated systems on human behaviour already during system development. Such models can also be integrated into the algorithms of automated vehicles, enabling them to better understand the human interaction partners’ behaviours. Further, the development of vehicle automation is much about technology (software and hardware), but the users will be humans and they will interact with humans both inside and outside of the vehicle. To be successful in the development of automated vehicles functionalities, research must be performed on a variety of aspects. Actually, a highly interdisciplinary team of researchers, bringing together expertise and background from various scientific fields related to traffic safety, human factors, human-machine interaction design and evaluation, automation, computational modelling, and artificial intelligence, is likely needed to consider the human-technology aspects of vehicle automation. Accordingly, SHAPE-IT has recruited fifteen PhD candidates (Early Stage Researchers – ESRs), that work together to facilitate this integration of automated vehicles into complex urban traffic by performing research to support the development of transparent, cooperative, accepted, trustworthy, and safe automated vehicles. With their (and their supervisors’) different scientific background, the candidates bring different theoretical concepts and methodological approaches to the project. This interdisciplinarity of the project team offers the unique possibility for each PhD candidate to address research questions from a broad perspective – including theories and methodological approaches of other interrelated disciplines. This is the main reason why SHAPE-IT has been funded by the European Commission’s Marie Skłodowska-Curie Innovative Training Network (ITN) program that is aimed to train early state researchers in multidisciplinary aspects of research including transferable skills. With the unique scope of SHAPE-IT, including the human-vehicle perspective, considering different road-users (inside and outside of the vehicle), addressing for example trust, transparency, and safety, and including a wide range of methodological approaches, the project members can substantially contribute to the development and deployment of safe and appreciated vehicle automation in the cities of the future. To achieve the goal of interdisciplinary research, it is necessary to provide the individual PhD candidate with a starting point, especially on the different and diverse methodological approaches of the different disciplines. The empirical, user-centred approach for the development and evaluation of innovative automated vehicle concepts is central to SHAPE- IT. This deliverable (D1.1 “Methodological Framework for Modelling and Empirical Approaches”) provides this starting point. That is, this document provides a broad overview of approaches and methodologies used and developed by the SHAPE-IT ESRs during their research. The SHAPE-IT PhD candidates, as well as other researchers and developers outside of SHAPE-IT, can use this document when searching for appropriate methodological approaches, or simply get a brief overview of research methodologies often employed in automated vehicle research. The first chapter of the deliverable shortly describes the major methodological approaches to collect data relevant for investigating road user behaviour. Each subchapter describes one approach, ranging from naturalistic driving studies to controlled experiments in driving simulators, with the goal to provide the unfamiliar reader with a broad overview of the approach, including its scope, the type of data collected, and its limitations. Each subchapter ends with recommendations for further reading – literature that provide much more detail and examples. The second chapter explains four different highly relevant tools for data collection, such as interviews, questionnaires, physiological measures, and as other current tools (the Wizard of Oz paradigm and Augmented and Virtual Reality). As in the first chapter this chapter provides the reader with information about advantages and disadvantages of the different tools and with proposed further readings. The third chapter deals with computational models of human/agent interaction and presents in four subchapters different modelling approaches, ranging from models based on psychological mechanisms, rule-based and artificial intelligence models to simulation models of traffic interaction. The fourth chapter is devoted to Requirements Engineering and the challenge of communicating knowledge (e.g., human factors) to developers of automated vehicles. When forming the SHAPE-IT proposal it was identified that there is a lack of communication of human factors knowledge about the highly technical development of automated vehicles. This is why it is highly important that the SHAPE-IT ESRs get training in requirement engineering. Regardless of the ESRs working in academia or industry after their studies it is important to learn how to communicate and disseminate the findings to engineers. The deliverable ends with the chapter “Method Champions”. Here the expertise and association of the different PhD candidates with the different topics are made explicit to facilitate and encourage networking between PhDs with special expertise and those seeking support, especially with regards to methodological questions.
|
|
| 8. |
- He, Yixu, et al.
(författare)
-
Exploring the design of reward functions in deep reinforcement learning-based vehicle velocity control algorithms
- 2024
-
Ingår i: Transportation Letters. - 1942-7867 .- 1942-7875. ; In Press
-
Tidskriftsartikel (refereegranskat)abstract
- The application of deep reinforcement learning (DRL) techniques in intelligent transportation systems garners significant attention. In this field, reward function design is a crucial factor for DRL performance. Current research predominantly relies on a trial-and-error approach for designing reward functions, lacking mathematical support and necessitating extensive empirical experimentation. Our research uses vehicle velocity control as a case study, build training and test sets, and develop a DRL framework for speed control. This framework examines both single-objective and multi-objective optimization in reward function designs. In single-objective optimization, we introduce “expected optimal velocity” as an optimization objective and analyze how different reward functions affect performance, providing a mathematical perspective on optimizing reward functions. In multi-objective optimization, we propose a reward function design paradigm and validate its effectiveness. Our findings offer a versatile framework and theoretical guidance for developing and optimizing reward functions in DRL, particularly for intelligent transportation systems.
|
|
| 9. |
- Li, Ying, et al.
(författare)
-
Deep knowledge distillation: A self-mutual learning framework for traffic prediction
- 2024
-
Ingår i: Expert Systems with Applications. - 0957-4174. ; 252
-
Tidskriftsartikel (refereegranskat)abstract
- Traffic flow prediction in spatio-temporal networks is a crucial aspect of Intelligent Transportation Systems (ITS). Existing traffic flow forecasting methods, particularly those utilizing graph neural networks, encounter limitations. When processing large-scale graph data, the depth of these models can restrict their ability to effectively capture complex relationships and patterns. Additionally, these methods often focus mainly on local neighborhood information, which can limit their capability to recognize and analyze global relationships and patterns within the graph data. Therefore, we proposed a deep knowledge distillation model, tailored to effectively capture spatio-temporal patterns in traffic flow prediction. This model incorporates a bidirectional random walk process on a directed graph, enabling it to effectively capture both spatial and temporal dependencies. Utilizing a blend of mutual learning and self-distillation, our approach enhances the detection of spatio-temporal relationships within traffic data and improves the feature perception ability at both local and global levels. We tested our model on two real-world datasets, achieving notable improvements in prediction accuracy, especially for predictions within a one-hour timeframe. In comparison to the baseline model, our proposed model achieved accuracy improvements of 0.19 and 0.18 on the respective datasets. These results highlight the success of using mutual learning and self-distillation to transfer knowledge effectively within and between models and to improve the model's capability in identifying and extracting features.
|
|
| 10. |
- Liu, Yang, 1991, et al.
(författare)
-
DeepTSP: Deep traffic state prediction model based on large-scale empirical data
- 2021
-
Ingår i: Communications in Transportation Research. - : Elsevier BV. - 2772-4247. ; 1
-
Tidskriftsartikel (refereegranskat)abstract
- Real-time traffic state (e.g., speed) prediction is an essential component for traffic control and management in an urban road network. How to build an effective large-scale traffic state prediction system is a challenging but highly valuable problem. This study focuses on the construction of an effective solution designed for spatio-temporal data to predict the traffic state of large-scale traffic systems. In this study, we first summarize the three challenges faced by large-scale traffic state prediction, i.e., scale, granularity, and sparsity. Based on the domain knowledge of traffic engineering, the propagation of traffic states along the road network is theoretically analyzed, which are elaborated in aspects of the temporal and spatial propagation of traffic state, traffic state experience replay, and multi-source data fusion. A deep learning architecture, termed as Deep Traffic State Prediction (DeepTSP), is therefore proposed to address the current challenges in traffic state prediction. Experiments demonstrate that the proposed DeepTSP model can effectively predict large-scale traffic states.
|
|
