deepdrive
deepdrive copied to clipboard
FR - tracking on other vehicles
Is it possible to get id and real world distance to other vehicles in the simulation?
Thanks Omer. It's available in Unreal, but we'd have to add it to the Python API. We're thinking of returning something like this:
locomotives:
types: pedestrians, vehicles, bicycles, motorcycles, animals, golf_carts
properties: id, type, position, rotation, height, width, length, velocity, rotational velocity, acceleration, up_vector, forward_vector, right_vector, transformation matrix, number_of_passengers, on_road, last_collision_time, position_relative_to_agent
There are a couple of ways we could do it. One would be to add an extra API call with a search radius, especially if this call is expensive. The other would be to return it every frame as we do with the existing data. In that case, we'd allow setting the search radius or perhaps only return objects visible to your cameras. I think the first crack at this should just add an API call.
Then for distance to the agent, you could use the length of the position_relative_to_agent
vector. We may not add all the fields above right away, so are there others besides distance that you think you will need?
Also, for folks doing computer vision, a property for whether the object was in the camera's view would be useful as well, perhaps using this. (Sorry this is slightly off topic, but wanted to centralize the discussion around locomotive objects)
Thanks, sounds perfect.