Trunk two-way image semantic segmentation method for scene understanding of mobile robot in complex environment

Abstract

The invention belongs to the field of image processing, and discloses a trunk two-way image semantic segmentation method, which specifically comprises the following steps: inputting an image into a semantic segmentation model for feature extraction to obtain an initial feature map; after the initial feature map is processed by a residual network, performing primary semantic fusion on the initial feature map and the initial feature map subjected to down-sampling and residual network processing to obtain a primary fusion high-resolution feature map and a primary fusion low-resolution feature map; performing residual network processing on the primary fusion high-resolution feature map, and performing secondary semantic fusion on the primary fusion high-resolution feature map and a primary fusion low-resolution feature map subjected to down-sampling and residual network processing; and obtaining a third-level fusion high-resolution feature map and a third-level fusion low-resolution feature map; processing the three-level fused high-resolution feature map through a residual network; carrying out third semantic fusion on the three-level fused high-resolution feature map and the three-level fusion low-resolution feature map subjected to down-sampling and residual network processing; and obtaining a five-level fusion feature map; and carrying out up-sampling on the five-level fusion feature map through a decoder to obtain an image semantic segmentation result.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a trunk two-way image semantic segmentation method for scene understanding of mobile robots in complex environments. Background technique [0002] For mobile robots, scene understanding is the core technology for realizing real intelligence. Its scene understanding ability depends on the high-precision semantic segmentation algorithm for scene analysis. A service robot with scene understanding ability has the ability of scene semantic segmentation. After moving the base and high-precision robotic arm, it can further realize advanced tasks such as autonomous navigation, object delivery, and indoor security. [0003] Scene understanding has been paid attention to in the early research work related to computer vision, but it has been unable to find a more effective solution. There are many difficulties in the research of scene understanding, such as: how to obtain the robu...

AI Technical Summary

Problems solved by technology

[0003] Scene understanding has been paid attention to in the early research work related to computer vision, but it has been unable to find a more effective solution

There are many difficulties in the research of scene understanding, such as: how to obtain the robust features of the target object when the target object in the scene is affected by translation, rotation, illumination or distortion, etc., in order to achieve better segmentation effect, research Personnel usually design complex structures for semantic segmentation models based on deep learning to improve segmentation accuracy, such as ASPP modules, etc., but complex structures usually slow down the running speed of the model. In order to improve the running speed of the model, many lightweight semantic segmentation models It was proposed, but the accuracy of the lightweight model has a certain gap compared with the precise model, the structure is usually special, it is difficult to improve or may need to be pre-trained on the ImageNet dataset after improvement, the equipment requirements are high, and the training is time-consuming and laborious

Images