Spatial Target Vision Measurement and Precision Compensation Based on Soft Computing

摘要

On the basis of geometrical relationship between three-dimensional spatial target system and double-theodolites, the measurement models of observing angle and spatial coordinates together with distance are established according to the principles of the space rendezvous and docking technology. In fact, due to the angles' trigonometric functions' nonlinearities and theodolite's inner characteristics, observing angle readings of double-theodolites produce errors at certain position. The reasons of such errors are analyzed, and the idea is provided of taking errors as nonlinear components, building up the Neural Network (NN) to simulate the nonlinear mapping between observing angles and distances, optimizing its weights, and regarding the outputs of NN as compensated term. Simulation curves imply weights of NN intermediate layer influence the final compensation precision. In experiment, NN with optimized weights is applied in the processing of measured data; the results of which certificate such idea adaptively compensate system errors' influences in binocular stereo vision.