Visual semantic relation detection method based on RGB-D image

Abstract

The invention discloses a visual semantic relation detection method based on an RGB-D image, and the method comprises the steps: firstly extracting an RGB feature vector of the image, estimating a depth map of the image, extracting a corresponding depth feature vector, and enabling the RGB feature vector and the depth feature vector to be fused; and inputting the fused feature vectors into a visual semantic relationship classifier, the visual semantic relationship classifier being designed to process the RGB feature vectors and the depth feature vectors at the same time and weaken the depth feature vector part to obtain a visual semantic relationship detection result. The detection precision and the generalization ability of the visual semantic relationship in a complex visual scene can beimproved.

Description

technical field [0001] The invention relates to deep learning and computer vision, in particular to a visual semantic relationship detection method in visual scene understanding. Background technique [0002] In the face of a large amount of image information, if the computer is to truly understand the content of the visual scene, it is not only necessary to detect each object in the image, but also to extract the interactive relationship between each object. Due to the complex visual scene in the image, traditional semantic relationship detection methods have problems such as low accuracy and insufficient generalization ability when extracting the visual semantic relationship. [0003] RGB-D images include RGB information and depth (Depth) information, which contains more information than traditional RGB images. At the same time, the fusion of RGB information and depth information in semantic relationship detection is more in line with the law of human cognition . The RGB...

AI Technical Summary

Problems solved by technology

Due to the complex visual scene in the image, the traditional semantic relationship detection method has problems such as low accuracy and insufficient generalization ability when extracting the visual semantic relationship.

Images