Spatial non-cooperative target attitude evaluation method with image scale transformation

Abstract

The invention discloses a spatial non-cooperative target attitude evaluation method with image scale transformation, which comprises non-cooperative target boundary extraction, image scale transformation, a modular depth separable convolutional network and three-axis attitude output, and is characterized in that an algorithm comprises two parts of training and detection, the training process mainly utilizes attitude image data marked on the ground,. Related parameters of the algorithm network are trained by using a gradient descent algorithm. in the detection process, the trained parameters are utilized to evaluate the three-axis attitude corresponding to the non-cooperative target in an image for the input image. The invention gets rid of the limitation on the configuration of the non-cooperative target in the posture recognition process, extracts the deep features of images step by step through the modular deep convolutional network, and solves the problems that the feature points are difficult to extract stably and the illumination adaptability is poor in a conventional method.

Description

technical field [0001] The invention relates to a non-cooperative target pose evaluation method, in particular to the non-cooperative target pose evaluation based on image features, and belongs to the field of image processing and pattern recognition. Background technique [0002] The attitude recognition of space non-cooperative targets is the premise of controlling and serving them at close range. Therefore, the recognition of non-cooperative target pose is an important research direction in the space field. The attitude assessment of non-cooperative targets based on visual images is mainly to match the local feature points of the image with the reference model to obtain the corresponding rotation matrix, and then use the least square method to solve the overdetermined equations to obtain the corresponding attitude angle. [0003] Usually, the difficulty in solving non-cooperative target poses based on monocular vision images lies in the extraction of feature points. Sol...

AI Technical Summary

Problems solved by technology

[0004] The problem solved by the technology of the present invention is: in order to overcome the problems of difficult robust extraction of image features and poor adaptability to illumination and distance in the gesture recognition process of spatial non-cooperative targets, the present invention proposes a spatial non-cooperative target with its own image scale transformation The attitude evaluation method solves the problem of feature point extraction and illumination adaptability through the introduction of modular convolutional network; through the image scale transformation module, it solves the problem of algorithm adaptability under different working distance conditions

Images