Outdoor road scene semantic segmentation method and device, electronic equipment and storage medium

Abstract

The invention discloses an outdoor road scene semantic segmentation method, which comprises the following steps: collecting RGB image information and depth information to obtain an RGB image and a depth image; creating a convolutional network architecture and a deconvolutional network architecture; inputting the current RGB image and the depth image into a convolutional network architecture to extract feature points layer by layer, fusing the feature points of the RGB image of each layer and the feature points of the depth image to serve as fused feature points, and outputting the fused feature points to the deconvolutional network architecture; and enabling the deconvolutional network architecture to perform reverse operation on the fusion feature points of each layer, performing optimization processing and then outputting a semantic segmentation image. According to the method, the segmentation is carried out through multi-dimensional fusion features of different layers, and the problems of wrong segmentation, cavities and the like caused by factors such as scenes, the illumination and weather during road segmentation are well solved; and finally, back-end optimization is performed on the segmentation image to enable edge segmentation to be accurate.

Description

technical field [0001] The invention relates to image processing technology, in particular to a method, device, electronic equipment and storage medium for semantic segmentation of outdoor road scenes. Background technique [0002] At present, the real-time semantic segmentation network generally uses a lightweight classification backbone network to extract features from RGB, and then uses FCN, UNet and other network structures for semantic segmentation. Outdoor main segmentation: road, people, motorcycles, bicycles, cars, trucks, buses, houses, sky, vegetation, etc. The real-time semantic segmentation network generally has the following problems: due to different scenes (various ground textures), lighting differences, weather and other factors, many problems such as mis-segmentation and holes are caused. Contents of the invention [0003] In order to overcome the deficiencies of the prior art, one of the objectives of the present invention is to provide a method for sema...

AI Technical Summary

Problems solved by technology

[0003] In order to overcome the deficiencies of the prior art, one of the purposes of the present invention is to provide a method for semantic segmentation of outdoor road scenes, which can solve the problems of wrong segmentation and holes in the semantic segmentation network of the prior art

[0004] The second object of the present invention is to provide an outdoor road scene semantic segmentation device, which can solve the problems of wrong segmentation and holes in the semantic segmentation network of the prior art

Images