ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question
0

Suggestions for centimeter-accuracy HD map creation

asked 2021-05-20 03:12:25 -0600

congphase gravatar image

updated 2021-05-23 04:01:13 -0600

I have successfully created working semantic map by using JOSM and draw on the Bing Imagery/Mapbox Sattelite,.. for Autoware.Auto's AVP Demo. But to my understanding, these are free services, the GNSS coordinates are therefore of dozens-of-meter-level accuracy. I have a car, equipped with quality GNSS sensors, cameras and lidar. If I want to have centimeter-level, or at least few-meters-level accuracy, what is the best procedure to achieve that with my own hardware?

Thank you for any of your advice.

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
1

answered 2021-05-30 13:37:56 -0600

Josh Whitley gravatar image

My recommendation is to create a geo-referenced point cloud and use a top-down view of that as the visual reference instead of the aerial imagery. Every point doesn't need to be geo-referenced as long as the origin is with fairly high accuracy (in both location and orientation). Unfortunately, I can't help much beyond that as the Foundation has always had mapping services donated or paid for them in the past.

edit flag offensive delete link more

Comments

Thank you @Josh Whitley,

I understand your idea. It's what I'll be doing too.

Just one thing to be clearer, the "origin" that you said, is the origin of the point cloud after being georeferenced right? And, how do we specify the origin of a point cloud? From the moment the whole point cloud is mapped successfully into a big one, i.e before the georeference step?

congphase gravatar image congphase  ( 2021-05-31 03:00:43 -0600 )edit
1

The origin that is created in the PCD file is the point at 0, 0, 0 prior to georeferencing and is coincident with the location in the map at which you started recording the map.

Josh Whitley gravatar image Josh Whitley  ( 2021-05-31 06:43:12 -0600 )edit

So, the reason for us to not have to georeference every points is that we just need some helpful ones (the origin is one of them) to get the transformation for all, including every other points. And that is enough, right?


If that assumption is true, then how many points do we need to georeference? And which points to choose to provide the best transformation?

congphase gravatar image congphase  ( 2021-05-31 21:20:36 -0600 )edit

Question Tools

2 followers

Stats

Asked: 2021-05-20 03:12:25 -0600

Seen: 286 times

Last updated: May 30 '21