Indoor object pose estimation method based on depth estimation and intra-class average shape

Abstract

The invention discloses an indoor object pose estimation method based on depth estimation and intra-class average shape, and the method comprises the steps: 1), obtaining basic data, including indoor scene RGB image data and indoor object three-dimensional model historical data; 2) estimating the depth of an indoor scene by using data, performing object segmentation and classification on the RGB image, calculating an intra-class average shape of a corresponding class of an object, and reconstructing normalized space coordinates of the object by combining a depth estimation result and the intra-class average shape; and 3) performing similarity transformation according to the normalized space coordinates and the depth map to obtain an indoor object pose estimation result. The defects of a current instance level object pose estimation method are overcome, pose estimation can be carried out on different instances of the same kind of objects, in addition, only RGB images are needed, depth images do not need to be obtained, and the problem that the depth images are difficult to obtain is solved.

Description

technical field [0001] The invention relates to the technical field of interior decoration design automation, in particular to a method for estimating the pose of indoor objects based on depth estimation and intra-class average shape. Background technique [0002] With the development of social economy and the improvement of people's quality of life, people began to pay more attention to the beauty and quality of life, and the reasonable and comfortable layout of indoor household objects can make the living environment more beautiful and greatly improve people's life. Quality and satisfy people's pursuit of a better life. In today's digital age, people can use some existing design software to virtualize the design of the home scene and get the corresponding visualization results, and carry out the final decoration arrangement according to the generated design drawings. Indoor object placement refers to selecting appropriate objects from the object database and placing them ...

AI Technical Summary

Problems solved by technology

There are also some works that have studied the pose estimation method of small-scale objects, but these works require obtaining the depth map of the indoor scene in advance. Compared with the RGB image, the acquisition of the depth image is more difficult and imprecise.

How to estimate the class-level pose of objects from RGB images only is a very challenging problem

Images