A5100421776- Autonomous Vehicle Technology and Safety
- Traffic and Road Safety
- Traffic control and management
- Reinforcement Learning in Robotics
- Advanced Neural Network Applications
- Vehicular Ad Hoc Networks (VANETs)
- Traffic Prediction and Management Techniques
- Robotics and Sensor-Based Localization
- Evacuation and Crowd Dynamics
- Vehicle Dynamics and Control Systems
- Video Surveillance and Tracking Methods
- Advanced Vision and Imaging
- Video Analysis and Summarization
- Cooperative Studies and Economics
- Fuel Cells and Related Materials
- Robotic Path Planning Algorithms
- Business and Economic Development
- Color perception and design
- Human-Automation Interaction and Safety
- Industrial Vision Systems and Defect Detection
- Safety and Risk Management
- Image Processing Techniques and Applications
- Multimodal Machine Learning Applications
- Machine Learning and ELM
- Human Pose and Action Recognition
Hunan University
2023-2025
Tsinghua University
2012-2024
Hebei University of Technology
2024
Wuxi Institute of Technology
2023-2024
Delft University of Technology
2023
Air University
2023
University Town of Shenzhen
2023
Shanghai Artificial Intelligence Laboratory
2022
Chinese Academy of Sciences
2019
Institute of Acoustics
2019
Acquiring large amounts of high-quality real sonar data for object detection autonomous underwater vehicles (AUVs) is challenging. Synthetic can be an alternative, but it hard to generate diverse using traditional generative models when are limited. This study proposes a novel style transfer method, i.e., the multigranular feature alignment cycle-consistent adversarial network (CycleGAN), images leveraging remote sensing images, which alleviate dependence on data. Specifically, we add...
Pedestrian behavior prediction is essential to enable safe and efficient driving of intelligent vehicles on urban traffic environment. This article presents a novel framework for pedestrian trajectory prediction, which integrates Dynamic Bayesian network Sequence-to-Sequence model through an adaptive online weighting method. utilizes environmental features kinematic information infer the pedestrian's motion intentions probabilistic reasoning. views predictions as sequence generation tasks,...
High-definition (HD) maps can provide an accurate representation of the road ahead and information on surrounding environment, which is key to success autonomous driving. Open-pit mine a typical unstructured environment with dynamic varying terrain, posing huge challenges environmental mapping. This study proposes HD map construction update system for driving in open-pit mines. First, we analyze elements characteristics According types, uses, frequency elements, establish multimodal...
Multiagent reinforcement learning (RL) training is usually difficult and time-consuming due to mutual interference among agents. Safety concerns make an already process even harder. This study proposes a safe adaptive policy transfer RL approach for multiagent cooperative control. Specifically, pioneer follower off-policy (PFOPT) method presented help agents acquire knowledge experience from single well-trained agent. Notably, the designed can both representation sample provided by in...
This study introduces a novel approach to neural rendering, specifically tailored for adversarial camouflage, within an extensive 3D rendering framework. Our method, named FPA, goes beyond traditional techniques by faithfully simulating lighting conditions and material variations, ensuring nuanced realistic representation of textures on target. To achieve this, we employ generative that learns patterns from diffusion model. involves incorporating specially designed loss covert constraint...
Motion forecasting is an important component in autonomous driving systems. One of the most challenging problems motion interactive trajectory prediction, whose goal to jointly forecasts future trajectories interacting agents. To this end, we present a large-scale prediction dataset named INT2 for INTeractive at INTersections. includes 612,000 scenes, each lasting 1 minute, containing up 10,200 hours data. The agent are auto-labeled by high-performance offline temporal detection and fusion...
Reinforcement learning (RL) has gained wide attention, but its implementation in autonomous vehicles is still limited by insufficient sample efficiency and heavy training costs. The of RL agents influenced the dimension state space, which can be partitioned to reduce complexity sampling computation. This study proposes a hierarchical clustering-based grouping reinforcement (HCSG-RL) method for switching decision vehicles. First, we partition base space into groups generate tree groups. Then,...
Well-trained visual object detectors are generally confronted with a severe performance decline when deployed in novel driving scenario due to the impact of domain shift. Despite excellent improvements unsupervised adaptive detection achieved by adversarial training, those approaches fail capture transfer core underlying holistic scenes. To solve this problem, we propose progressive critical region framework for cross-domain detection. Specifically, exploit potential foreground mining (PFM)...
Scene text editing aims to replace the source with target while preserving original background. Its practical applications span various domains, such as data generation and privacy protection, highlighting its increasing importance in recent years. In this study, we propose a novel Text Editing network Explicitly-decoupled transfer Minimized background reconstruction, called STEEM. Unlike existing methods that usually fuse style, content, background, our approach focuses on decoupling style...
Methods for 3D lane detection have been recently proposed to address the issue of inaccurate layouts in many autonomous driving scenarios (uphill/downhill, bump, etc.). Previous work struggled complex cases due their simple designs spatial transformation between front view and bird's eye (BEV) lack a realistic dataset. Towards these issues, we present PersFormer: an end-to-end monocular detector with novel Transformer-based feature module. Our model generates BEV features by attending...
In comparison to conventional traffic designs, shared spaces promote a more pleasant urban environment with slower motorized movement, smoother traffic, and less congestion. the foreseeable future, will be populated mixture of autonomous vehicles (AVs) vulnerable road users (VRUs) like pedestrians cyclists. However, driver-less AV lacks way communicate VRUs when they have reach an agreement negotiation, which brings new challenges safety smoothness traffic. To find feasible solution...