Semantic segmentation of vehicle vision based on two-branch Enet network

Authors:

Abstract
Semantic segmentation has important research significance in environment perception in unmanned field. By improving the Enet model, adding spatial branches, turning the original model into a two-branch structure, cropping the original structure as semantic branches, and using the attention mechanism to fuse the spatial and semantic branch feature maps to retain the image spatial information and improve the grid effect. The improved algorithm is tested on dataset Cityscapes and dataset BDD100K, and the average cross-merge ratio is improved by 2.75% and 2.41%, respectively. The implementation of the two-branch Enet network makes it easier to obtain information on the drivable area of the vehicle, obstacles ahead, etc.
Funding Information
  • National Natural Science Foundation of China
  • Beijing Union University

This publication has 12 references indexed in Scilit: