38 3 Vol. 38, No. 3 2009 6 Act a Geo da etic a et Carto graphic a Sinic a J un.,2009 :100121595 (2009) 0320216207 1,2, 1 1., 430079 ; 2., 430079 A Novel Method of Multi2ima ge Matching Using Ima ge and Sp ace Synthe sis Information YUAN Xiuxiao 1,2, MING Ya ng 1 1. School of Remote Sensing a nd Information Engineering, Wuhan University, Wuha n 430049, China ; 2. St at e Key La boratory of In2 formation Engineering in Surveying, Mapping a nd Remote Sensing, Wuha n University, Wuha n 430049, China Abstract : A novel method is introduced for multi2ima ge matching by synthesizing ima ge and sp ace information. Four level pyramid ima ges are generated firstly according to the rule that the next level pyramid is generated from the previous level using the avera ge gray values of the 3 3 pixels, and the first level pyramid is generated from the original ima ge. The ini2 tial horizont al p arallaxes between the reference ima ge and each searching ima ge are calculated in the highest pyramid level. Then corresponding ima ge point s are searched in each stereo pairs from the third level pyramid ima ges, and the matching results in all stereo pairs are integrated into the object sp ace, by which the mismatched ima ge points can be e2 liminated and more accurate spatial information can be obt ained for the subsequent pyramid ima ge matching. The method ba sed on correlation coefficient with geometric constraint s and global relaxation optimization is introduced in the process of matching. Finally, the fea sibility of the method proposed in this paper is verified by the experiments using a set of actu2 al digital aerial ima ges with big overlap. Compared with the tra ditional matching with two ima ges, the accuracy of DSM generated using our method shows that multi2ima ge matching can eliminate the mismatched point s more effectively and improve the success ratio of matching more significantly. Key words : multi2image matching ; DSM ; image matching based on correlation coefficient with geometric constraints ; re2 laxation matching ; accuracy : 3 3 4, ; 3,,, ;, DSM,, :; (DSM) ;;; : P231 :A : (40771176) ; (40721001) (Digital Surface Model, DSM), DSM 3,,,,, DSM 3, Gabet DSM [1 ],,,Okutomi Kanade
3 : 217 [2 ], Canu [3 ] SNCC ( Sum of Normalized Cross2correlation) DSM [4 ],, ;,,Zit nich Webb [ 5 ],, DSM :,, ; ;,,, 1 DSM 3, 1 DSM Fig. 1 Workflow of DSM generation based on multiple image matching 1 1. 1 [627 ], [1 ],,,
218 J une 2009 Vol. 38 No. 3 AGCS http :/ / xb. sinomap s. com 1. 2 [8210 ] 3 3 4 3 3,,, 1. 3 4, Z ( x1, y1 ) ( X, Y, Z), ( X, Y, Z) ( x2, y2 ), (x = x2 - x1,y = y2 - y1 ),, 100200 1. 4, 10 10, Fgrstner [4 ], ( X0, Y0, Z0 ), Z, p 2 pmin ( Z0 - Z) pmax ( Z0 +Z), p,p,12,, ( 0. 5), 1. 5,,,,, Ii m I j ( j = 1,2,, m), j ( j = 1,2,, m),ii ω I j P( i, j) = j k = m k = 1 k (1), 8 Ik I i, Il ( l = 1,2,, n), Ii ω I j I k ω Il C( i, j ; k, l) = T e - (p2 x +p2 y ) / (2),p x = ( x j - x l ) - ( x l - x k ) li I k x,p y = ( y j - y l ) - ( y l - y k ) Ii I k y p x, ; T, 1 25 8 Q ( n) ( i, j), P ( n+1) ( i, j) = m P ( n) ( i, j) Q ( n) ( i, j) s = 1 P ( n) ( i, s) Q ( n) ( i, s) (3),Q ( n) (i, j) = P ( n) (i, j) (c0 + c1) P ( n) ( k, I k <( I ) i i =1 m l) C(i, j ; k, l),c0,c1, n, ( Ii) Ii 8 (3), 0. 9,, 1. 6,,,,,, [11 ], n,,
3 : 219, : v x = - a11 X - a12y - a13z - ( x - x 0 ) v y = - a21x - a22y - a23z - ( y - y 0 ) [11 ] (4), (4) 2 n + 2,3,, [12 ],,, 0,, 3, ( 1 ),, ;, 2 Fig. 2 Overview of the experimental images 2 2. 1, 5 ( 2 ) DSM 9m, 0. 1 m,80 % POS Wu2 CA PS 176 GPS [ 13 ], 6 2 526 3 45 : 7. 1 cm, 4. 8 cm 2,, (a ) (b ),, TIN, 3 DSM 3, DSM (a ), ; ( b ) 3 DSM Fig. 3 DSM generated by multiple image matching and interpolation 2. 2 DSM DSM, 176, DSM 2 5, 34,, 1 DSM,2 1 DSM Tab. 1 Comparisons of DSM accuracy generated by differ2 ent image combinations / ( %) / m / m 33/ 34 68. 9 0. 224 0. 571 33/ 34/ 35 76. 7 0. 183 0. 366 33/ 34/ 35/ 36 77. 3 0. 170 0. 355 32/ 33/ 34/ 35/ 36 79. 2 0. 153 0. 326
220 J une 2009 Vol. 38 No. 3 AGCS http :/ / xb. sinomap s. com 2 DSM Tab. 2 Comparisons of the interpolated elevation of check points from the DSMs generated by different image combinations / m 33/ 34 33/ 34/ 35 33/ 34/ 35/ 36 32/ 33/ 34/ 35/ 36 5021000 504. 621 510. 080-5. 459 504. 546 0. 074 504. 546 0. 074 504. 556 0. 065 5010615 503. 762 503. 331 0. 431 503. 334 0. 428 503. 334 0. 428 503. 772-0. 010 5041681 506. 013 505. 748 0. 265 505. 939 0. 074 505. 972 0. 041 505. 979 0. 034 5051819 503. 286 503. 466 0. 180 503. 466 0. 180 503. 257 0. 029 503. 259 0. 027 5051737 507. 770 507. 560 0. 210 507. 560 0. 210 507. 652 0. 118 507. 652 0. 118 4061362 506. 636 506. 374 0. 262 506. 493 0. 143 506. 560 0. 076 506. 560 0. 076 5031463 513. 997 512. 040 1. 958 514. 209-0. 212 514. 209 0. 212 514. 152-0. 155 5031464 514. 078 514. 328-0. 251 514. 225-0. 147 514. 225 0. 147 514. 137-0. 059 6031331 505. 426 505. 767-0. 340 505. 285 0. 141 505. 285 0. 141 505. 285 0. 141 4030965 502. 211 502. 566-0. 356 502. 234-0. 023 502. 234-0. 023 502. 229-0. 018 5061595 511. 021 510. 535 0. 486 510. 842 0. 179 510. 815 0. 206 510. 815 0. 206 5041681 506. 013 505. 748 0. 265 505. 939 0. 074 505. 972 0. 041 505. 979 0. 034 4041203 503. 855 504. 133-0. 279 503. 904-0. 049 503. 824 0. 031 503. 824 0. 031 1 2 : 1.,,68. 9 % 79. 2 %,, ;,, 2., DSM,3 DSM (0. 571-0. 366) / 0. 571 = 35. 9 %,4 DSM (0. 571-0. 355) / 0. 571 = 37. 8 %,5 DSM ( 0. 571-0. 326 ) / 0. 571 = 42. 9 % 5 DSM, 0. 326 m, 3. 3 pixels,3,, 2 5021000 5031463, 33/ 34 5021000-5. 459 m,,,5 0. 065 m,5031463 1. 958 m 5-0. 155 m 3.,3 DSM, 3,1,,,,,, DSM, DSM,, DSM,,, DSM, 2 5 DSM,4 3 4,2 DSM,,,5 DSM,, DSM
3 : 221 3 DSM Tab. 3 Comparison results of different matching methods CPU / ( %) / m / m / h I 73. 7 0. 157 0. 340 1. 91 II 80. 2 0. 177 0. 327 1. 88 79. 2 0. 153 0. 326 1. 10 4 DSM Fig. 4 Enlarged views of local DSM generated by the method proposed in this paper 2. 3, DSM I [1 ],, DSM ; II [ 4 ],SNCC, 3 3 DSM, 5 DSM 3 5 : 1. II 80 %,DSM,I 2. I,,,, II, SNCC,, DSM 5 DSM (A ), I,II 3. II,, SNCC SNCC,, [4 ]4,,3,,, II,CPU 5 3 DSM Fig. 5 Partial results of DSM generated by three methods
222 J une 2009 Vol. 38 No. 3 AGCS http :/ / xb. sinomap s. com 3,,, DSM,,,,,,,, 3,DSM,,, DSM : [ 1 ] GABET L, GIRAUDON G, RENOUARD L. Automatic Generation of High Resolution Urban Zone Digital Eleva2 tion Models [J ]. International Journal of Photogrammetry and Remote Sensing, 1997, 52 (1) : 33247. [ 2 ] O KU TOMI M, KANADE T. A Multiple2baseline Stereo [J ]. IEEE Transaction on Pattern Analysis and Machine Intelligence, 1993, 15 (4) :3532363. [ 3 ] CANU D, A YACHE N, SIRA T J A. Accurate and Robust Stereovision with a Large Number of Aerial Images [ C]/ / Proceedings of SPIE on Satellite Remote Sensing II. Paris : SPIE t he International Society for Optical Engineering, 1995 (2579) :1522160. [ 4 ] ZHAN G Li. Automatic Digital Surface Model (DSM) Genera2 tion from Linear Array Images [D]. Zurich : ETH, 2005. [ 5 ] ZITNICK C, WEBB J A. Multi2baseline Stereo Using Sur2 face Extraction [ R]. Pittsburgh : Carnegie Mellon Univer2 sity, 1996. [ 6 ] PA TERA KI M, BAL TSAVIAS E. Adaptive Multi2image Matching Algorithm for the Airborne Digital Sensor ADS40 [ DB/ OL ]. [ 2002 ]. http :/ / citeseerx. ist. psu. edu/ view2 doc/ summary? doi = 10. 1. 1. 8. 5242. [ 7 ] J IAN G Wanshou. Multiple Aerial Image Matching and Au2 tomatic Building Detection [ D ]. Wuhan : Wuhan Universi2 ty, 2004. (. [ D]. :, 2004. ) [ 8 ] ZHAN G Zuxun, ZHAN G Jianqing. Principles of Digital Photogrammetry [ M ]. Wuhan : Wuhan University Press, 2002. (,. [ M ]. :, 2002. ) [ 9 ] MALLA T S. A Theory for Multiresolution Signal Decom2 position : The Wavelet Representation [J ]. IEEE Transac2 tion on Pattern Analysis and Machine Intelligence, 1989, 11 (7) : 6742693. [ 10 ] PETER J B, EDWARD H A. The Laplcian Pyramid as a Compact Image Code [J ]. IEEE Transaction on Commu2 nication, 1983, 31 (4) : 5322540. [ 11 ] L I Deren, ZHEN G Zhaobao. Principles of Analytical Pho2 togrammetry [ M]. Beijing :Surveying and Mapping Press, 1992. (,. [ M ]. :, 1992. ) [ 12 ] L I Deren, YUAN Xiuxiao. Error Processing and Reliabili2 ty Theory [ M]. Wuhan : Wuhan University Press, 2002. (,. [ M ]. :, 2002. ) [ 13 ] YUAN Xiuxiao. A Novel Met hod of Systematic Error Compensation for a Position and Orientation System [J ]. Progress in Natural Science, 2008, 18 (8) :9532963. : 2008203227 : 2008211217 ( :) : (19632),,,, E2mail :yxxqxhyw @public. wh. hb. cn First a uthor : YUAN Xiuxiao (19632), male, PhD, Pro2 fe ssor, PhD supervisor majors in the re se arch a nd e du2 c ation in remote sensing (RS), glo bal po sitioning sys2 tem ( GPS) a nd their inte gra tio n, theorie s a nd metho ds for high precisio n photo gra mmetric po sitioning, GPS/ IMU2supported a erotria ngulation, geometric proce ssing of high2re solutio n satellite ima gery, etc.. E2mail :yxxqxhyw @public. wh. hb. cn