lane detection, deep learning, autonomous driving, CNN, RNN, LSTM, GRU, lane localization, lane fitting, ego lane, end-to-end, vanishing point, segmentation, FCN, regression, classification
2. Outline
• CurveLane-NAS: Unifying Lane-Sensitive Architecture Search and Adaptive Point
Blending
• Heatmap-based Vanishing Point boosts Lane Detection
• LDNet: End-to-End Lane Detection Approach using a Dynamic Vision Sensor
• RONELD: Robust Neural Network Output Enhancement for Active Lane Detection
• Keep your Eyes on the Lane: Real-time Attention-guided Lane Detection
• End-to-end Lane Shape Prediction with Transformers
• End-to-End Deep Learning of Lane Detection and Path Prediction for Real-Time
Autonomous Driving
3. CurveLane-NAS: Unifying Lane-Sensitive
Architecture Search and Adaptive Point Blending
• Lane-sensitive architecture search framework, CurveLane-NAS;
• Three parts:
• a) feature fusion search module, for multi-level hierarchy features to
fuse local and global context;
• b) flexible backbone search module, semantic feature extractor;
• c) adaptive point blending module, search for a multi-level post
processing refinement strategy to combine multi-scale head prediction.
• CurveLanes:around 150K images, 90% for curved lanes (~ 135K
images).
10. Heatmap-based Vanishing Point boosts Lane Detection
• vanishing point as structure info;
• ERFNet for segmentation;
• Heatmap regression for VP prediction.
14. LDNet: End-to-End Lane Detection Approach
using a Dynamic Vision Sensor
• Lane Detection using dynamic vision sensor (LDNet)
Comparison of Normal Camera and Event Camera
18. RONELD: Robust Neural Network Output
Enhancement for Active Lane Detection
• Real-time robust neural network output enhancement for active
lane detection (RONELD)
21. Keep your Eyes on the Lane: Real-time
Attention-guided Lane Detection
LaneATT: Anchor-based lane detection, similar to Line-CNN.
22. Keep your Eyes on the Lane: Real-time
Attention-guided Lane Detection
23. Keep your Eyes on the Lane: Real-time
Attention-guided Lane Detection
24. End-to-end Lane Shape Prediction with Transformers
• Directly estimate the lane shape model with Transformers (contextual info).
25. End-to-end Lane Shape Prediction with Transformers
The backbone: a reduced transformer network, several feed-forward networks (FFNs) for
parameter predictions, and Hungarian bipartite matching Loss.
28. End-to-End Deep Learning of Lane Detection and
Path Prediction for Real-Time Autonomous Driving
• Three-task convolutional
neural network (3TCNN);
• Include a bounding box
regression branch, a Hu
moments regression branch
and a classification branch;
• Lateral offset-path prediction.
TORCS