Gerhard Koenig, Wolfgang Nickel, Juergen Storl

Transcription

1 DIGITAL S TEREOPHOTOGRAMMETRY EXPERIENCE WITH AN EXPERIMENTAL SYSTEM Gerhard Koenig, Wolfgang Nickel, Juergen Storl Technical Univerity of Berlin Photogramrnetry and Cartography StraBe de 17. Juni 135 D-1000 Berlin 12, FRG Commiion II ABSTRACT An experimental Dig al Stereophotogrammetric Sytem (DSS), which i baed on a commercially available image proceing ytem, ha been et up at the Technical Univerity of Berlin. Any kind of tereocopic imagery, uch a canning electron microcope data, cloe range imagery, aerial photograph and atellite image data, can be proceed tereophotogrammmetrically. Thi paper report on the practical experience concerning photogrammetric proceing, epecially of aerial photograph. Due to hardware retriction, o far only ubection of image frame can be handled in reaonable time. The performance of the ytem a well a the accuracy that could be achieved are dicued and compared to the reult obtained with an Analytical Plotter. Baed on the practical experience with the experimental DSS concluion are drawn concerning the deign and architecture of an operational ytem. An outlook i given to further ytem development, conidering new hardware component. INTRODUCTION At the ISPRS meeting of Commiion II, WGII/2 in Baltimore 1986, the author reported on an experimental Digital Stereophotogrammetric Sytem (DSS) which ha been et up at the Technical Univerity of Berlin. Thi ytem then wa uitable for automatic proceing of imagery reduced in ize (512x512 pixel), uch a digital canning electron microcope image. Baed on the exiting hardware decribed in (KOENIG, NICKEL, STORL, 1986) the author have implemented algorithm for the photogrammetric proceing of large data et, epecially of digitized aerial photograph. DATA ACQUISITION For converting the analogue aerial photograph into rater data, a color canner of HELL company, Chromagraph DC 360, i ued. u

2 Thi ytem can handle tranparency up to the format of 65x51 cm 2 a ize which i limited by the diameter of the rotating canning drum. The aerial photograph can be digitized in different reolution. According to the experience of the author, a reolution of 25 ~ or 400 line/em i ufficient for tereophotogrammetric purpoe. The advantage of higher reolution i too mall to juti it cot. Thi mean that a digitized tereo pair take about 180 MB of dik torage. Analyi of the canned imagery ha indicated that the Chromagraph can with high quality and very little geometric ditortion. The radiometric reproduction qual i good o that only few preproceing tep are neceary for image improvement that the quality of the photograph i ufficient. ORIENTATION The or ation i done in two tep, i.e. relative orientation and abolute orientation. After relat ation the image are reampled to the normal cae of tereophotogrammetry. Thi lead to a more rational automatic evaluation and enable tereo viion of the whole model. To overlook the cene both image are miniaturized to the ize of one creen. A 23 em x 23 em aer photograph canned with a xel pac of 25 micron lead to about 9000x9000 pixel. 20x20 p l are and thu the whole cene can be ualized at one t The ampled miniaturization are held permanent in two of the four image memorie. The cene part to be actual proceed are defined poitioning a 25x25 l window and are loaded the two rema memorie. For each (at leat two of) the fiducial mark region are way. Fiducial mark matching then i performed by a (via an artificial template or via binarization and calculation). Image inner orientation i known and lead to the er of an affine tranformation which compute image coordinate from pixel (i.e. canner) coordinate. RELATIVE ENTATION From up to any given number homologou point can be matched interact ly for relative orientation. Thi can be done either defining template in the firt and earch window in the econd image or by uing ignalized object control point, if there are any. can be earched in one image and matched to the other one or be earched independently in both image. Image coordinate again are given by the aff tranformation mentioned above. After thi calculation reampling i carried out for the whole tereo area (normal u bil interpolat ) to eliminate y-parallaxe. Thu the image are at hand for tereocopic viion by anag c mean and proceing along horizontal lar l.

3 ABSOLUTE ORIENTAT If the ground control point have not been meaured for relat ation thi ha to be done now. U the re (i.e. the relatively o ientated ) intead of the inal image, tereo c view of the earth urface i the time thi i realized in an anag ic way. A three dimen 1 float mark i generated in the memory and can be moved by a moue in x- and y the creen. Z direct motion of the float formed pre two funct button of the the DSS can be handled like a convent Control po i. model have which lead to an error le than 0.5 accurracy on the tization rate of the 1. a uff accuracy of abolute orientation i meaur leat ix ground control po. U abolute orientat reulted in tandard plan deviat of 0. 7 m for both the ic Plotter KERN DSR 11 and the DSS. The tandard he were 0.04 m and 0.06 m, repect cale ~ 1: 5000). 3-D PROCESSING Automated evaluat of a D i a main feature of the DSS. At preent i performed f homologue po in two in an in ially preregular grid in object pace. algorithm ue the Vertical L Locu (VLL method BETHEL 1986) to calculate approximate hei. The f correlation i done a leat quare ithm 1983) In order to ave time there i no ubpixel interpolation calculated earch and the center p 1 of the template i et to the nearet ne Thu l reult doe not repreent a object pace Normally h ae al failure can not be been developed Thee ria t 20% orrelati ome cr e have in correlat - Dep e exit lation may the tandard defined 1 ry laxe caued of a parallax i et for the actual a pre-

4 the correlation coeffic i normal 0 8) uch reult are autocorrelat funct in the large lax fference occur between a threhold value calculat the of the a cent a. corre- proce ou wrong reearch failure and the texture i ued a above reult in The detected on the on gro error and to correct ly. Furthermore, he allowed amount and the of calculation. A or the DTM on different cate that the trat e to correlat reult are very and 40% o excluded - even reliable correlation reult. Accori necce to de more den than the one to have a u c number of po for lation If number i too mal a be calculated DTM 11 be ued to evaluate hei for re-tart the automatic proce to r amount of homologue Orientation t and correlation need about one econd and po evaluation o te 1987) were 0.02 % of the t about the ame h Anal time 1 ection ze FUTURE DEVE the t oftware tool author have performance of te fferent ource. The ne ie will de of an operational em conit new hardware. Thi em will be de around a powerful work-

5 tation which for performance purpoe hould be baed on a Reduced truction ~ornputer SC) that ha a relat ly imple architecture In contrat to conventional, a RISC conit of relative few ruction name e ch are ued mot ly. Mot of thee intruction could be a e and have a xed format for have a large ter et and are compiler which lead to a performance rate around 10 MIPS In ion to uch a pe al hardware nent, a de in, are neceary for ett up a real t irne phot ric ern The ue of parallel proce architecture i neceary in orde to obtain the performance needed ric application tate of the could be claified a follow: Intruction Mult le Data SIMD) proceor, where each proceor execute the ame intruct at the ame time, but on fferent data element; - Multiple Intruct Mult (MIMD) proceor which can be egmented in Mult er, where each proceor ha it own attached memory and each execute own program code, and Mult MIMD proceor hare a common memory - and S olic where each proceor a ion of an algor ation in a p parallel fa. All are zed to and at the ame t Modern image comput ytem make ue of thee arch ecture, for example PIXAR (SIMD), or AT&T~ Pixel (MIMD), which i build around up to 82 Digital Signal Proceor, rated at lomflop each. Thee ytem offer a computational power which i neceary for and image proceing applicat. are equ a memory - up to 48MB are common - and in mot cae could be configured parallel tranfer dik, which accomodate the torage of data et, uch a aerial, and the tranfer rate needed for fat proceing language 1 C-callable for un the tak to thee An advanced te ric worktation with h formance hould well a colour mon 1 tereo di ue of 1 1 hutter techno-logy. The ue of a le pecta le with polarizing ae make a comfortable and flicker-free view of the tereocopic image poible and eem to be the optimal olution.

6 CONCLUSIONS Experience ha hown that proceing of tereocopic imagery by ue of conventional ytem architecture and photogrammetric tandard algorithm lead to atifactory reult concerning the accuracy. Compared with conventional plotting in an Analytical Plotter the algorithm implemented in the DSS for automated evaluation are too time-conuming. At preent a new photogrammetric worktation baed on tate-of-the-art technology i under deign, which will olve thee problem. REFERENCES ACKERMANN F., 1983: High Preciion Digital Image Correlation. Proceeding of the 39th Photogrammetric Week, Stuttgart, pp ALBERTZ J., 1986: Digitale Bildverarbeitung in der Nahphotogrammetrie - Neue Moglichkeiten und Aufgaben. Bildmeung und Luftbildween, Vol. 54, pp BETHEL J., : The DSRll Image Correlator. ASP/ ACSM Convention, Wahington D.C., 6 pp. GRUEN A., 1987: Toward Real-Time-Photogrammetry. Invited Paper, 41th Photogrammetric Week, Stuttgart, 33 pp. KOENIG G., NICKEL W., STORL J., 1986: Proceing of Scann Electron Microcope Imagery in a Digital Stereophotogrammet c Sytem (DSS). International Archive of Photogrammetry and Remote Sening, Vol. 26-2, pp KUPFER G., 1987: Volle geometriche Sytemkalibrierung metricher Luftbildkammern Da Tetfeld Brecherpitze. Bildmeung und Luftbildween, Vol. 55, pp