Point-Painting is the method used to fuse the semantic segmentation results based on RBG images to raw Lidar point cloud.
This is an implementation of a pipeline to combine the data from LiDAR and Camera to obtain semantic painted point cloud. DeepLabV3+ network has been used on KITTI-360 dataset to do semantic segmentation for the RGB images and then using these RGB semantics and extrinsics between camera and LiDAR, each 3D point in the point cloud has been labelled.
Download the stereo camera images and Velodyne sensor data from any of the 9 sequences of the KITTI-360 dataset and store it in the Code/Data folder as specified. The dataset can be downloaded from here.
Doppler ICP is a novel algorithm for point cloud registration for range sensors capable of measuring per-return instantaneous radial velocity. Existing variants of ICP that solely rely on geometry or other features generally fail to estimate the motion of the sensor correctly in scenarios that have non-distinctive features and/or repetitive geometric structures such as hallways, tunnels, highways, and bridges. from https://github.com/aevainc/Doppler-ICP
-
To do semantic segmentation of RGB images, get the pretrained model of DeepLabV3plus from here.
-
Go to the parent folder of this repo, that is, LiDAR_3D_semantics and enter the command:
python3 Wrapper.py --path **path_to_dataset_folder**
