Method and Apparatus for Instance Segmentation Based Object Size Estimation from Single Indoor Image

摘要

A method and apparatus for estimating object size from a single indoor image based on instance segmentation are presented. The method for estimating the size of an object from a single indoor image based on instance division proposed in the present invention includes performing preprocessing including Gaussian filtering and noise removal on the original image, and detecting Canny edges on the preprocessed image and Hough transform to the vanishing point. Finding candidates and detecting the vanishing point through grid search for the vanishing point candidates, the instance segmentation step to find the mask of the reference object through mask R-CNN model learning for the preprocessed image, and the size from the target object to measure the size The method includes selecting points for measurement and calculating the size of the target object using points for measuring the size of the selected target object, a mask of a reference object, and an intersection rate with respect to the vanishing point.