Skip to content

Latest commit

 

History

History
37 lines (32 loc) · 1.71 KB

lightweight_semantic_mesh.md

File metadata and controls

37 lines (32 loc) · 1.71 KB

Lightweight Semantic Mesh Mapping for Autonomous Vehicles

Author: Herb

Year: 2021

Notes:

  • 3D semantic mesh by fusing VO, dense depth estimation with deep learning and semantic with deep learning
  • Voxel and surfel based methods are computationnaly expensive and require a lot of storage (e.g. SurfelMeshing, On fast surface reconstruction methods for large and noisy point clouds, Real-time Scalable Dense Surfel Mapping)
  • Meshing based on sparse set of kp: Incremental reconstruction of urban environments by edge-points delaunay triangulation, Real-time CPU-based large-scale 3D mesh reconstruction

Method:

  • ORB-SLAM2 as VO system
  • PSPNet for 2D semantic segmentation
  • Perform Constrained Delaunay Triangulation + Delaunay Refinement
  • Estimate the covariance of each vertex in the camera frame using 2D and depth uncertainties
  • Need to perform probabilistic vertex matching 3D to 2D, as it is not using the local map of ORBSLAM2.
  • An incremental mesh update is performed for each object:
    • Compute a 3D Mesh from the current active Mesh
    • Conflict detection (overlapping semantics)
    • Update probabilistic vertices using fast Covariance Intersection

Exp:

  • Use GA-Net to predict depth from stereo or struct2depth to predict depth from monocular
  • Compare to TSDF reconstruction that is the semantic component of KIMERA
  • Presents qualitative results on images and videos
  • Evaluation metric: semantic chamfer distance with point cloud generated from mesh with a density of 2500 pts/m^2: $$ e_{g \rightarrow \mathcal{R}}=\min {r \in \mathcal{R}, c_g=c_r}|r-g| \quad e{r \rightarrow \mathcal{G}}=\min _{g \in \mathcal{G}, c_r=c_g}|g-r| . $$
  • Evaluate on KITTI and their own dataset
  • Fastest method