Environment semantic mapping method based on deep convolutional neural network

Abstract

The invention provides an environmental semantic mapping method based on a deep convolutional neural network, and the method can build an environmental map containing object category information by combining the advantages of deep learning in the aspect of scene recognition with the autonomous positioning advantages of an SLAM technology. In particular, ORB-is utilized Carrying out key frame screening and inter-frame pose estimation on the input image sequence by the SLAM; Carrying out two-dimensional semantic segmentation by utilizing an improved method based on Deeplab image segmentation; Introducing an upper sampling convolutional layer behind the last layer of the convolutional network; and using the depth information as a threshold signal to control selection of different convolutionkernels, aligning the segmented image and the depth map, and constructing a three-dimensional dense semantic map by using a spatial corresponding relationship between adjacent key frames. According tothe scheme, the image segmentation precision can be improved, and higher composition efficiency is achieved.

Description

technical field [0001] The invention relates to the fields of digital image processing and computer vision, in particular to an environment semantic mapping method based on a deep convolutional neural network. Background technique [0002] Accurately obtaining environmental information is a key link for mobile robots to perform tasks autonomously. When the robot is moving, it can form a description of the surrounding environment through SLAM technology, that is, an environmental map. However, traditional SLAM composition only considers geometric data, and cannot obtain the category and type of objects in the map. The information provided is insufficient and the distinction of features is relatively weak. Semantic information includes object category, target detection and semantic segmentation, etc., which can understand the content of the scene and help the robot perform tasks in a goal-oriented manner. Therefore, the combination of the two is an inevitable requirement. ...

AI Technical Summary

Problems solved by technology

There are also proposals in the industry to use dense Elastic Fusion to build a 3D environment map, and use a deconvolutional neural network for image segmentation. This method uses dense Elastic Fusion to build maps, which is relatively time-consuming and resource-intensive.

Images