| In the field of lunar exploration engineering,accurate camera parameters are one of the key factors for high-precision navigation and positioning of patrol.In the orbit environment,the number of high-precision control points is small,and the straight line information in the navigation images can be used as an additional constraint to assist in the high-precision on-orbit calibration of the stereo navigation camera.Therefore,it is of great significance to realize the automatic extraction and matching of the straight line in the solar panel area in the navigation images.However,because there is no atmosphere in the lunar environment,there is a very strong illumination contrast in the navigation images containing solar panels,which eventually leads to the fact that only a small number of straight lines with single direction can be extracted by traditional methods,while the straight lines in the areas with weak gradient information are completely missing,which leads to the lack of auxiliary information in the subsequent camera calibration process and reduces the calibration accuracy of navigation cameras.Therefore,around the problem that it is difficult to extract straight lines in dark gray areas in navigation images,this paper focuses on the following aspects:(1)Aiming at that problem that traditional image denoising method need to manually determine parameters,an adaptive nonlocal mean filtering method is proposed,and several group of contrast experiments of noise filtering are carried out,which rely on Mean squared error(MSE),Signal to noise ratio(SNR),Structual similarity,SSIM)and other indicators verify the effectiveness and feasibility of the proposed filtering method.Experiments show that compared with traditional median filtering,the MSE of the proposed filtering method is reduced by 47.37%,SNR is increased by 13.47%,and SSIM is increased by 12.89%.(2)Aiming at the problem that the feature pixels in the image cannot be processed centrally,an adaptive edge extraction method between partitions is proposed.By optimizing the gradient template,the gradient of the image R,G,B and gray 4 channels is statistically calculated to ensure the loss of weak pixel gradient information,and the statistics are performed according to the gradient direction of the actual neighborhood of the pixel,and the confidence interval is arranged according to its distribution law and Gaussian distribution.The properties of the image are adaptively partitioned,and pixel sets in different directions and different degrees are obtained.Count the 3×3 neighborhood brightness of each pixel,and resample its neighborhood brightness according to the actual gradient direction distribution of its center pixel,and use the average brightness of 9 pixels as its final brightness to obtain a more realistic pixel brightness.feature.According to the relationship between the gradient and the brightness,the gradient threshold is solved for the pixels in each interval,so as to extract the weak edge.Experiments show that the MSE of the edge extraction method proposed in this paper is reduced by 1.93% and 3.52% on average compared with the other two methods,and the SNR is increased by 0.44% and 0.67% on average.(3)Aiming at the problem that edge extraction results can’t form a straight line,an adaptive straight line extraction method is proposed.According to the obtained weak edge results,the edge chains are sorted,the seed point tracking method is used to detect the straight line according to the gradient consistency principle,the straight line is fitted by the least square principle,the false straight line is eliminated by Helmholtz principle,and the final straight line is obtained by length and angle constraints.The experiment shows that compared with LSD and EDlines,this method can correctly extract long lines with the lifting rate of 36.34% and 25.07%,and the average lifting rate of short lines is 38.28% and 24.95%.(4)Through the subjective interpretation judgment method and the line segment matching method,the straight line extraction results of this method,LSD and EDlines are compared and analyzed,and the straight line extraction and correct matching number are analyzed and evaluated.The correct straight line matching rate of the image with gray overall brightness increased by 58.44% and 49.95% on average.After comparing several groups of images taken by Chang ’e-3 navigation camera in different stations,the qualitative experiment proves that the proposed method can extract more real weak line information than the traditional line extraction method,and the comprehensive performance of the proposed line extraction method is proved by the quantitative evaluation experiment of line matching.In addition,the matching result of line extraction is used to calibrate the self-calibration camera with additional constraints for verification.Compared with LSD and EDlines,the calibration accuracy of the straight line segment distribution results is improved by 2.88% and 1.12% respectively,which indicates that the straight line segment extraction method proposed in this paper has a better extraction effect and can provide a certain value reference for navigation camera on-orbit calibration.There are 48 figures,13 tables and 72 references in this paper. |
PDF Full Text Request