The Autoware Foundation
-
|
We try object detection in the room. |
Beta Was this translation helpful? Give feedback.
Replies: 0 comments 1 reply
-
|
The solution is to crop the input pointcloud excluding the room ceiling here. Centerpoint model uses a PointPillars-based network to inference. Apollo model is bird eye view based and the ceiling pointcloud bothers the detection. |
Beta Was this translation helpful? Give feedback.
The solution is to crop the input pointcloud excluding the room ceiling here.
Centerpoint model uses a PointPillars-based network to inference.
The pillar unit depends on the top pointcloud including room ceiling, which bothers the detection.
See japanese reference and PointPillars paper.
Apollo model is bird eye view based and the ceiling pointcloud bothers the detection.
https://github.com/ApolloAuto/apollo/blob/master/docs/specs/3d_obstacle_perception.md