search

3dv-casia / PanoVLM

PanoVLM: Low-Cost and Accurate Panoramic Vision and LiDAR Fused Mapping
129 stars 11 forks source link

The point cloud is very sparse after coloring #3

Closed Yellowshuohahaha closed 11 months ago

Yellowshuohahaha commented 11 months ago

Why？ This is quite different from the number of point clouds originally collected in the dataset. image

DiantaoTu commented 11 months ago

ColorizeLidarMap only processes lidar points with range [near, far]. And the final lidar cloud is fused every k cloud. By default, near=1.5, far=35, k=5. You can change the settings in the code. Decreasing k to zero can fuse every lidar cloud but lead to a significantly large file size. https://github.com/3dv-casia/PanoVLM/blob/84ae30e22db0b1d454693ba1a152fed350e2d23f/main.cpp#L545-L546

Yellowshuohahaha commented 11 months ago

ColorizeLidarMap only processes lidar points with range [near, far]. And the final lidar cloud is fused every k cloud. By default, near=1.5, far=35, k=5. You can change the settings in the code. Decreasing k to zero can fuse every lidar cloud but lead to a significantly large file size.

https://github.com/3dv-casia/PanoVLM/blob/84ae30e22db0b1d454693ba1a152fed350e2d23f/main.cpp#L545-L546

That's work, thank you very much! The overall point cloud became much denser, but the color looked blurry in some places. Is there any test data for other scenarios? image

DiantaoTu commented 11 months ago

The blur is because of the misalignment between images and lidar clouds. The two sensors capture data with different viewing angles and some occlusion may occur, leading to wrong colorizing. This phenomenon is more obvious in indoor environment. We are currently compiling the dataset of outdoor scenes and it will be available as we finish.

Yellowshuohahaha commented 11 months ago

@DiantaoTu Thank you for your answer!

Yellowshuohahaha commented 11 months ago

The blur is because of the misalignment between images and lidar clouds. The two sensors capture data with different viewing angles and some occlusion may occur, leading to wrong colorizing. This phenomenon is more obvious in indoor environment. We are currently compiling the dataset of outdoor scenes and it will be available as we finish.

Misalignment refers to the timestamp of the image and lidaris out of sync or their pose is not aligned?

DiantaoTu commented 11 months ago

Mainly because the camera and lidar are not strictly synchronized, the time offset is about 100ms to 200ms. Even the camera and lidar poses are correct, directly projecting the lidar clouds to images can have a certain degree of misalignment.

Yellowshuohahaha commented 11 months ago

Get it! Thanks.