Research & Projects
Detailed technical overview of ongoing and completed research projects. Each project includes methodology, architecture, results, and research contributions.
Hybrid DeepLabV3+ & U-Net Lane Segmentation
Click to collapse details
Objective
Develop a robust lane detection system combining semantic segmentation architectures for real-time autonomous driving in complex road environments.
Methodology
Hybrid architecture merging DeepLabV3+ atrous spatial pyramid pooling (ASPP) with U-Net encoder-decoder skip connections. Multi-scale feature extraction enables precise pixel-level lane boundary detection even under occlusion and varying illumination.
Model Architecture
Performance Metrics
Research Figures & Outputs
Raw camera feed from vehicle
Predicted lane boundaries
Annotated lane labels
Key Results
Achieved 96.2% accuracy on CULane benchmark with real-time inference at 30+ FPS. Outperforms standalone DeepLabV3+ by 3.8% and U-Net by 5.1% in adverse conditions.
Research Contribution
Novel hybrid architecture improving lane detection in adverse weather, low-light, and occluded conditions. Demonstrates feasibility of real-time deployment on embedded vehicle platforms.
Bio-Inspired Hybrid Control (ACO + PSO)
Click to expand details
AI-Driven Microgrid Energy Management
Click to expand details
Autonomous Vehicle Simulation Environment
Click to expand details
Research Environment
State-of-the-art facilities and tools powering our research.
LiDAR Sensors
3D point cloud generation
NVIDIA GPU
A100 for deep learning training
CARLA Simulator
Autonomous driving simulation
ROS
Robot Operating System
TensorFlow
Deep learning framework
MATLAB
Control system simulation
OpenCV
Computer vision processing
Python
Primary research language
PyTorch
Neural network research
Docker
Containerized environments
GitHub
Version control & collab
Linux
Development OS
Research Network
Intelligent Connected Vehicle Lab
A state-of-the-art laboratory at HUAT focused on next-generation autonomous driving algorithms and connected infrastructure. Collaborating with industry partners on real-world deployment challenges.
Key Collaborators
Industry Partner: Dongfeng Motors • Institution: HUAT
Research Questions
Active discussions and questions posed on ResearchGate.
How to Address Real-Time Inference and Generalization Challenges in Lane Detection Models for Autonomous Vehicles?
I am working on lane detection systems for autonomous vehicles, leveraging a hybrid model that combines DeepLabv3+ and U-Net. While these models deliver high accuracy, I am facing challenges related to real-time inference and generalization in diverse environmental conditions.
How Can Hybrid Models (DeepLabv3++ & U-Net) Help Autonomous Vehicles With Lane Detection?
I'm now utilizing a hybrid model that combines DeepLabv3++ and U-Net architectures to work on lane detection for autonomous cars. Although the outcomes are encouraging, I'm looking for methods to improve the model's functionality even more, particularly in difficult situations.
What are the most significant challenges and promising solutions for ensuring the ethical and unbiased decision-making of autonomous vehicles?
Focuses on a critical aspect: This question delves into the crucial ethical and societal implications of AI in autonomous vehicles, a core concern for researchers and the public.