August 2020
tl;dr: Map buildings and roads as polygon.
Identify keypoints first, then starting with one arbitrary one vertex, connect them according to the Left/Right hand rule (or Maze solving algorithm), then there is one unique way to define the graph.
The following work are focused on road network discovery and are NOT focused on HD maps.
- DeepRoadMapper: semantic segmentation
- RoadTracer: like an DRL agent
- PolyMapper: iterate every vertices of a closed polygon
Polyline loss and DAGMapper focuses on HD mapping tasks with lane-level information.
Road network extraction is still very helpful for routing purposes, but lack the fine detail and accuracy needed for a safe localization and motion planning of an autonomous car.
- Find RoI with RPN
- Identify keypoints
- Connect keypoints with RNN (Conv-LSTM)
- Summary of technical details
- Questions and notes on how to improve/revise the current work