Yong Li’s research while affiliated with Beijing University of Technology and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (17)


Fig. 1. Roadside scenario. (a) shows the roadside scene captured by LiDAR; (b) displays the roadside scene captured by a camera; and (c) depicts a bird's-eye view diagram of the roadside scene, indicating the positions of LiDAR, cameras, and other infrastructure.
Fig. 5. Depth Features Fusion Module. The RGB image is first input into a modified ResNet architecture, which extracts multi-scale RGB features. These features are then processed by the Multi-Scale Feature Modulation (MSFM) module to estimate depth at corresponding scales, generating multi-scale depth features. Both RGB and depth features are subsequently refined through convolutional layers, resulting in integrated RGB-Depth features enriched with detailed depth information.
Fig. 6. Examples of calibration results on the DAIR-V2X-I dataset are illustrated as follows: (a) shows the original image plane; (b) displays the projection of mis-calibrated point clouds; (c) depicts the corrected projection using our method predicted extrinsic parameters; and (d) shows the projection with the ground truth parameters.
Fig. 7. Visualization results compared with other state-of-the-art algorithms at 5 • /0.5m range.
Fig. 8. Visualization results compared with other state-of-the-art algorithms at 10 • /0.25m range.

+2

RLCFormer: Automatic Roadside LiDAR-Camera Calibration Framework with Transformer
  • Article
  • Full-text available

September 2024

·

10 Reads

Heliyon

Rui Tian

·

Xuefeng Bao

·

Yunli Chen

·

[...]

·

Yong Li

LiDAR-Camera fusion is pivotal for perceiving and understanding complex traffic environments, particularly valuable in autonomous driving and traffic monitoring. Traditional calibration algorithms, primarily designed for onboard sensors, are inadequate for roadside setups where sensors are positioned higher and more dispersed. To address this challenge, we introduce the RLCFormer, a Transformer-based network specifically tailored for precise calibration of roadside sensors. This method innovatively integrates depth and RGB features, utilizing correlation layers and a Transformer decoder to accurately match features across modalities. Evaluated on the DAIR-V2X-I Roadside 3D detection dataset, the RLCFormer achieves an average translation error of 3.3187 cm and a rotation error of 0.0469°, surpassing existing methods. Our approach significantly enhances scene representation and calibration precision, offering a robust solution for roadside sensor calibration and advancing the state of the art in sensor fusion technology.

Download


Construction of a large-scale maritime element semantic schema based on knowledge graph models for unmanned automated decision-making

June 2024

·

48 Reads

·

1 Citation

In maritime logistics optimization, considerable research efforts are focused on the extraction of deep behavioral characteristics from comprehensive shipping data to discern patterns in maritime vessel behavior. The effective linkage of these characteristics with maritime infrastructure, such as berths, is critical for the enhancement of ship navigation systems. This endeavor is paramount not only as a research focus within maritime information science but also for the progression of intelligent maritime systems. Traditional methodologies have primarily emphasized the analysis of navigational paths of vessels without an extensive consideration of the geographical dynamics between ships and port infrastructure. However, the introduction of knowledge graphs has enabled the integration of disparate data sources, facilitating new insights that propel the development of intelligent maritime systems. This manuscript presents a novel framework using knowledge graph technology for profound analysis of maritime data. Utilizing automatic identification system (AIS) data alongside spatial information from port facilities, the framework forms semantic triplet connections among ships, anchorages, berths, and waterways. This enables the semantic modeling of maritime behaviors, offering precise identification of ships through their diverse semantic information. Moreover, by exploiting the semantic relations between ships and berths, a reverse semantic knowledge graph for berths is constructed, which is specifically tailored to ship type, size, and category. The manuscript critically evaluates a range of graph embedding techniques, dimensionality reduction methods, and classification strategies through experimental frameworks to determine the most efficacious methodologies. The findings reveal that the maritime knowledge graph significantly enhances the semantic understanding of unmanned maritime equipment, thereby improving decision-making capabilities. Additionally, it establishes a semantic foundation for the development of expansive maritime models, illustrating the potential of knowledge graph technology in advancing intelligent maritime systems.


biSAMNet: A Novel Approach in Maritime Data Completion Using Deep Learning and NLP Techniques

May 2024

·

30 Reads

·

1 Citation

Journal of Marine Science and Engineering

In the extensive monitoring of maritime traffic, maritime management frequently encounters incomplete automatic identification system (AIS) data. This deficiency poses significant challenges to safety management, requiring effective methods to infer corresponding ship information. We tackle this issue using a classification approach. Due to the absence of a fixed road network at sea unlike on land, raw trajectories are difficult to convert and cannot be directly fed into neural networks. We devised a latitude–longitude gridding encoding strategy capable of transforming continuous latitude–longitude data into discrete grid points. Simultaneously, we employed a compression algorithm to further extract significant grid points, thereby shortening the encoding sequence. Utilizing natural language processing techniques, we integrate the Word2vec word embedding approach with our novel biLSTM self-attention chunk-max pooling net (biSAMNet) model, enhancing the classification of vessel trajectories. This method classifies targets into ship types and ship lengths within static information. Employing the Taiwan Strait as a case study and benchmarking against CNN, RNN, and methods based on the attention mechanism, our findings underscore our model’s superiority. The biSAMNet achieves an impressive trajectory classification F1 score of 0.94 in the ship category dataset using only five-dimensional word embeddings. Additionally, through ablation experiments, the effectiveness of the Word2vec pre-trained embedding layer is highlighted. This study introduces a novel method for handling ship trajectory data, addressing the challenge of obtaining ship static information when AIS data are unreliable.



biSAMNet: A Novel Approach in Maritime Data Completion Using Deep Learning and NLP Techniques

April 2024

·

88 Reads

In the extensive monitoring of maritime traffic, maritime management frequently encounters incomplete automatic identification system (AIS) data, particularly in critical static fields like vessel types. This shortfall presents substantial challenges in safety management, necessitating robust methods for data completion. Distinct from road traffic, where road width imposes constraints, maritime vessels of identical classifications often choose significantly divergent routes, deviating by several nautical miles. Sole dependence on trajectory clustering is insufficient. Addressing this issue, we segments the maritime area into spatiotemporal grids, transforming vessel paths into sequences of grid encodings. Utilizing natural language processing techniques, we integrate the Word2vec word embedding approach with our novel biLSTM self-attention chunk-max pooling net (biSAMNet) model, enhancing the classification of vessel trajectories. This method is crucial for determining static details like vessel types. Employing the Taiwan Strait as a case study and benchmarking against CNN, RNN, and methods based on the attention mechanism, our findings underscore our model’s superiority. The biSAMNet achieves an impressive vessel classification F1 score of 0.94, using only 5-dimensional word embeddings, highlighting the effectiveness of Word2vec pre-trained embedding layers. This research introduces a novel paradigm for processing vessel trajectory data, greatly enhancing the accuracy of filling in incomplete vessel type details.



Maritime greenhouse gas emission estimation and forecasting through AIS data analytics: a case study of Tianjin port in the context of sustainable development

December 2023

·

100 Reads

·

5 Citations

The escalating greenhouse gas (GHG) emissions from maritime trade present a serious environmental and biological threat. With increasing emission reduction initiatives, such as the European Union’s incorporation of the maritime sector into the emissions trading system, both challenges and opportunities emerge for maritime transport and associated industries. To address these concerns, this study presents a model specifically designed for estimating and projecting the spatiotemporal GHG emission inventory of ships, particularly when dealing with incomplete automatic identification system datasets. In the computational aspect of the model, various data processing techniques are employed to rectify inaccuracies arising from incomplete or erroneous AIS data, including big data cleaning, ship trajectory aggregation, multi-source spatiotemporal data fusion and missing data complementation. Utilizing a bottom-up ship dynamic approach, the model generates a high-resolution GHG emission inventory. This inventory contains key attributes such as the types of ships emitting GHGs, the locations of these emissions, the time periods during which emissions occur, and emissions. For predictive analytics, the model utilizes temporal fusion transformers equipped with the attention mechanism to accurately forecast the critical emission parameters, including emission locations, time frames, and quantities. Focusing on the sea area around Tianjin port—a region characterized by high shipping activity—this study achieves fine-grained emission source tracking via detailed emission inventory calculations. Moreover, the prediction model achieves a promising loss function of approximately 0.15 under the optimal parameter configuration, obtaining a better result than recurrent neural network (RNN) and long short-term memory network (LSTM) in the comparative experiments. The proposed method allows for a comprehensive understanding of emission patterns across diverse vessel types under various operational conditions. Coupled with the prediction results, the study offers valuable theoretical and data-driven support for formulating emission reduction strategies and optimizing resource allocation, thereby contributing to sustainable maritime transformation.


Automatic Roadside Camera Calibration with Transformers

November 2023

·

135 Reads

·

2 Citations

Sensors

Previous camera self-calibration methods have exhibited certain notable shortcomings. On the one hand, they either exclusively emphasized scene cues or solely focused on vehicle-related cues, resulting in a lack of adaptability to diverse scenarios and a limited number of effective features. Furthermore, these methods either solely utilized geometric features within traffic scenes or exclusively extracted semantic information, failing to comprehensively consider both aspects. This limited the comprehensive feature extraction from scenes, ultimately leading to a decrease in calibration accuracy. Additionally, conventional vanishing point-based self-calibration methods often required the design of additional edge-background models and manual parameter tuning, thereby increasing operational complexity and the potential for errors. Given these observed limitations, and in order to address these challenges, we propose an innovative roadside camera self-calibration model based on the Transformer architecture. This model possesses a unique capability to simultaneously learn scene features and vehicle features within traffic scenarios while considering both geometric and semantic information. Through this approach, our model can overcome the constraints of prior methods, enhancing calibration accuracy and robustness while reducing operational complexity and the potential for errors. Our method outperforms existing approaches on both real-world dataset scenarios and publicly available datasets, demonstrating the effectiveness of our approach.


A Practical Large-Scale Roadside Multi-View Multi-Sensor Spatial Synchronization Framework for Intelligent Transportation Systems

November 2023

·

10 Reads

·

1 Citation

p>Spatial synchronization in roadside scenarios is essential for integrating data from multiple sensors at different locations. Current methods using cascading spatial transformation (CST) often lead to cumulative errors in large-scale deployments. Manual camera calibration is insufficient and requires extensive manual work, and existing methods are limited to controlled or single-view scenarios. To address these challenges, our research introduces a parallel spatial transformation (PST)-based framework for large-scale, multi-view, multi-sensor scenarios. PST parallelizes sensor coordinate system transformation, reducing cumulative errors. We incorporate deep learning for precise roadside monocular global localization, reducing manual work. Additionally, we use geolocation cues and an optimization algorithm for improved synchronization accuracy. Our framework has been tested in real-world scenarios, outperforming CST-based methods. It significantly enhances large-scale roadside multi-perspective, multi-sensor spatial synchronization, reducing deployment costs.</p


Citations (5)


... Finally, Data Transformation involves converting text data into a format more suitable for analysis, such as vector representation, using techniques such as TF-IDF, Word2Vec, and other word embeddings [27]. By applying these preprocessing steps, raw text data is transformed into a more structured and meaningful form, improving the performance of machine learning models and producing more accurate and effective analysis results [28]. ...

Reference:

Early Stopping on CNN-LSTM Development to Improve Classification Performance
biSAMNet: A Novel Approach in Maritime Data Completion Using Deep Learning and NLP Techniques

Journal of Marine Science and Engineering

... The intertwining of maritime activities with global trade has laid the foundation for unparalleled economic growth, with ports serving as crucial nodes in the international trade network. However, this economic boon comes with significant environmental costs: emissions from ships, particularly in marine and riverine areas, have become a potent source of air pollution, degrading air quality, accelerating global warming, and posing greater health risks to nearby populations [15,35]. The maritime sector, which contributes substantially to global emissions, finds itself in a dilemma-providing an essential service to international trade while simultaneously having a considerable environmental impact. ...

Maritime greenhouse gas emission estimation and forecasting through AIS data analytics: a case study of Tianjin port in the context of sustainable development

... By integrating the predicted calibration result T pred from the network with the initial calibration parameter T init , the extrinsic calibration parameters between the uncalibrated LiDAR and the camera are derived. These extrinsic calibration parameters are denoted as shown in Equation (13). ...

Automatic Roadside Camera Calibration with Transformers

Sensors

... Additionally, deep learning techniques have also been employed to predict traffic on major maritime routes. Li et al. (2023) utilized Spatiotemporal Graph Neural Networks (ST-GNNs) to effectively forecast maritime traffic trajectories in the waters surrounding the Netherlands. Sørensen, Heiselberg, and Heiselberg (2022) applied Bidirectional Long Short-Term Memory (BLSTM) and Mixture Density Network (MDN) models to project critical traffic patterns near Norway. ...

Research on Multi-Port Ship Traffic Prediction Method Based on Spatiotemporal Graph Neural Networks

Journal of Marine Science and Engineering

... The primary contributors to these emissions are ship's main engine, auxiliary engine, and boiler. According to relevant research works [40][41][42][43], the emission calculation method for ship engines is outlined in Eq. (2) and Eq. (3). ...

Research on the carbon emissions traceability inventory and multi-horizon prediction of ship carbon emissions: a case study of Tianjin Port