Spatial non-cooperative target relative pose estimation method based on deep learning

Abstract

The invention discloses a spatial non-cooperative target relative pose estimation method based on deep learning, and the method comprises the steps: making a data set of a non-cooperative target through software, and expanding the data set; designing a target detection network based on a convolutional neural network, judging whether the target is a non-cooperative target or not, and detecting a 2Dbounding box around the non-cooperative target; segmenting the non-cooperative target around the 2D bounding box, searching the center of the non-cooperative target based on Kirchhoff, estimating thedepth from the center of the non-cooperative target to the camera, and converting 2D pixel coordinates into 3D coordinates in combination with a depth value; designing a key point extraction network,extracting key points such as corner points and three-axis end points of the non-cooperative target, and performing regression on the key points to obtain a rotation quaternion to represent rotation;and finely tuning the estimated relative pose through iterative optimization. According to the method, the conditions of severe shielding, sensor noise, low estimation precision during object symmetry and the like can be processed, and meanwhile, the real-time speed requirement can be met.

Description

technical field [0001] The invention belongs to the field of spacecraft navigation, in particular to a method for estimating the relative pose of a space non-cooperative target based on deep learning. Background technique [0002] Attitude determination is to determine the attitude of the body with star sensors and gyroscopes. At present, most of the tasks faced are maintenance of failed spacecraft and capture of out-of-control spacecraft, and the research objects are non-cooperative targets in space. This kind of space non-cooperative target often rolls rapidly out of control in space, so it is necessary to obtain the position and attitude of the non-cooperative target in the case of unknown appearance characteristics, no response, and no identification. Collect images of non-cooperative targets through a monocular camera, provide a large number of real-time images, and then use the collected image information to estimate the structure and motion of the target. The existing...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the requirements of real-time and autonomy in non-cooperative target pose determination and the actual needs of low power consumption and low cost of micro-satellites, and overcome the shortcomings of traditional visual measurement methods that cannot cope with severe occlusion, changing illumination environments, and complex model structures , the present invention provides a method for estimating the relative pose of a spatial non-cooperative target based on deep learning. The method uses a synthetic image and an actual image taken by a monocular camera as input, and obtains a spatial non-cooperative target by designing a convolutional neural network. Target position and attitude, complete multiple space tasks including space capture, etc., and can realize real-time estimation of the attitude and position of space non-cooperative targets

Images