3D perception algorithm of unstructured environment based on point cloud enhanced pixel fusion
Guo Chen, Liming Wan, Jingjing He, Shuhao Jiang, Lin Song · 2026
Read 3D perception algorithm of unstructured environment based on point cloud enhanced pixel fusion by Guo Chen, Liming Wan, Jingjing He, Shuhao Jiang, Lin Song (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13562740.txt.
Description
Guo Chen and four colleagues fuse laser point clouds with camera images so robots can judge the three-dimensional world outside tidy settings. Streets and work sites shift constantly, and neither sensor suffices alone: laser returns carry depth but sparse detail, while photos are rich but flat. Their 2026 framework, PEPF-Net, first enriches each image pixel with the depth and reflection strength of nearby laser points. A vector-attention module then learns the geometry among the points, letting the two data kinds merge at feature level rather than by projection.
A layered transformer extracts image features in stages, and a variable-window attention step keeps computation local where detail matters. Trials on the KITTI driving benchmark, a testbed for such systems, placed the network ahead of advanced methods for 3D perception. The authors pitch the design for robots that must find, size, and track objects in uncontrolled surroundings.
Catalog record
| CREATOR | Guo Chen, Liming Wan, Jingjing He, Shuhao Jiang, Lin Song |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 80513 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |