A kinect-based object support relationship inference method

Abstract

The invention discloses a Kinect-based object supporting relationship inference method. The method comprises the following steps: segmenting an image by using a hierarchical segmentation algorithm based on an RGB-D image; building a scene structure level model, a scene supporting relationship model based on a structure level and a supporting relationship restraining rule; evaluating a scene supporting relationship by using a linear plan; inferring the stability of an object in a scene by using the scene supporting relationship. By adopting the Kinect-based object supporting relationship inference method, depth data acquired by Kinect is effectively utilized, the supporting relationship in the scene is extracted innovatively, and the vision and target identification work of a robot can be effectively assisted.

Description

technical field [0001] The invention relates to the field of scene analysis, in particular to a Kinect-based object support relationship inference method. Background technique [0002] With the development of camera acquisition equipment, image information including depth data has been able to be acquired. How to make good use of depth information to make it play a greater role in the fields of robot vision and object recognition has become a research hotspot. In recent years, a lot of research has been done on the scene analysis based on RGB-D data, but the analysis of the scene structure is still lacking, and the support relationship obtained through the scene structure can well assist robot vision and determine the stability of objects in the scene. Being able to provide novel recognition features for object recognition will surely become an important part of scene understanding in the future. [0003] Among the existing scene structure extraction methods, the main meth...

AI Technical Summary

Problems solved by technology

[0004] Focusing on the extraction of the overall structure information of the scene, it can only outline the simple overall three-dimensional information of the scene, but in the current situation where depth information can be collected, the overall structure information of the scene can no longer meet the needs of scene structure extraction

[0005] The extraction of the object bounding box in the scene can only extract the bounding box of a single object, ignoring the connection between different objects in the scene, which cannot reflect the connection between objects well

[0006] There are few researches on the existing support relationship extraction algorithm. In the existing technology, the extraction of the core scene constraint relationship in the scene is not comprehensive enough, resulting in a low accuracy rate of the algorithm.

Images