Target space positioning method

Abstract

The invention discloses a target space positioning method, which comprises the following steps of: simultaneously acquiring two target images with different visual angles, and obtaining a three-dimensional coordinate set of each pixel point in one image through binocular vision space positioning; based on instance segmentation, performing classification regression on a target in the image to obtain a target binary mask set; and based on the three-dimensional coordinate set of each pixel point and the target binary mask set, through pixel point coordinate mapping and fusion, obtaining a targetthree-dimensional coordinate in the image, and realizing target space positioning. Based on binocular vision space positioning, sparse three-dimensional coordinates used for describing the real scaleof a target and space positioning information are obtained. A deep learning method is adopted to perform monocular instance segmentation on a specific category of interested targets, and semantic attributes of pixels are accurately defined. Finally, based on coupling of the three-dimensional coordinates and the instance segmentation result, target space positioning is performed under the connection relationship of the pixel coordinates so that the sparsely dispersed three-dimensional coordinates are enabled to be dense and the positioning accuracy can be enhanced.

Description

technical field [0001] The invention belongs to the field of target space positioning, and more particularly relates to a target space positioning method. Background technique [0002] With the continuous development of production and life, the location information of the target has begun to be concerned by more and more fields. The spatial positioning of the target has a wide range of applications in many application scenarios, such as alarming of dangerous areas in the factory area, automatic driving obstacle prediction, and aerospace Position and attitude estimation, etc. [0003] The existing target space positioning methods mainly include hardware-assisted methods, traditional geometric methods, and depth estimation methods based on deep learning. The hardware-assisted method mainly refers to the calculation and positioning of the target through the analysis and calculation of the transmitted / received active signal with the help of radio frequency identification, parti...

AI Technical Summary

Problems solved by technology

[0005] The present invention provides a target space positioning method, which is used to solve the technical problem that the existing target space positioning method cannot overcome the inherent drawbacks due to the independent use of a single type of method such as geometric method and deep learning method, which leads to insufficient spatial positioning accuracy

Images