This is a big one. As far as I know, we have no code for this yet. Likely, we will need to break this into smaller issues as we refine the implementation.
Although a CNN option was originally tossed around as an idea, the difficulty in getting good training data pushed us back towards a more algorithmic OpenCV approach. Some things will be tricky since the lane edges are dashed and not continuous.
Ideally, the edges of the lane can be interpolated between neighboring dashes and published as "obstacles" so the avoidance method can easily interpret the lane constraints.
This is a big one. As far as I know, we have no code for this yet. Likely, we will need to break this into smaller issues as we refine the implementation.
Although a CNN option was originally tossed around as an idea, the difficulty in getting good training data pushed us back towards a more algorithmic OpenCV approach. Some things will be tricky since the lane edges are dashed and not continuous.
Ideally, the edges of the lane can be interpolated between neighboring dashes and published as "obstacles" so the avoidance method can easily interpret the lane constraints.