A5027090154- Video Surveillance and Tracking Methods
- Evacuation and Crowd Dynamics
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Autonomous Vehicle Technology and Safety
- Advanced Vision and Imaging
- Advanced Image and Video Retrieval Techniques
- Human Pose and Action Recognition
National University of Defense Technology
2025
Southeast University
2021-2023
Abstract 3D human pose and shape estimation is the foundation of analyzing motion. However, estimating accurate temporally consistent motion from a video remains challenge. By now, most video‐based methods for rely on unidirectional temporal features lack more comprehensive information. To solve this problem, we propose novel model “bidirectional feature recovery” (BTMR), which consists generator discriminator. The transformer‐based effectively captures forward reverse to enhance correlation...
It is essential for researchers to have a proper interpretation of remote sensing images (RSIs) and precise semantic labeling their component parts. Although FCN (Fully Convolutional Networks)-like deep convolutional network architectures been widely applied in the perception autonomous cars, there are still two challenges segmentation RSIs. The first identify details high-resolution with complex scenes solve class-mismatch issues; second capture edge objects finely without being confused by...
Abstract We propose a virtual crowd navigation approach based on deep reinforcement learning to improve the adaptability of crowds in an unknown and complex environment. To address problem local optimum or slow iteration even failure converge due sparse rewards environments, we integrate curiosity‐driven mechanism, key points acquisition path penalty method addition combining long short‐term memory networks, dynamic obstacle collision prediction with proximal policy optimization algorithms,...