HIGH-RESOLUTION WAVEFORM ACQUISITION AND ANALYSIS OF LASER PULSES

Transcription

1 HIGH-RESOLUTION WAVEFORM ACQUISITION AND ANALYSIS OF LASER PULSES B. Juzi a, J. Neulis a, U. Silla b a FGAN-FOM Research Insiue for Opronics and Paern Recogniion, 7675 Elingen, German - juzi@fom.fgan.de b Phoogrammer and Remoe Sensing, Technische Universiae Muenchen, 8090 Muenchen, German - silla@bv.um.de KEY WORDS: Urban, Waveform, Analsis, Laser scanning, LIDAR, Measuremen, Feaure. ABSTRACT: In his paper we describe invesigaions for digial recording of received laser pulses and a deailed analsis of he pulse waveform. An eperimenal ssem was consruced using an airborne laser scanning ssem and a receiver uni wih a high-resoluion sampling rae, which is above he sampling rae of curren commercial laser ssems. Capuring he emporal waveform b simulaneous scanning of he scene delivers a daa volume (space-ime cube). Such a daa cube of he received signals allows eploiing differen feaures wihou selecing a special feaure or a measuring echnique in advance. For he main invesigaion a ground based sensor plaform is used o measure he urban scene. Surface variaions wihin he spaial beam widh resul in special waveforms. We eplored he capabiliies of analzing he shape of he waveform for surface feaures in form of disance, roughness, and reflecance for each backscaered pulse. An ieraive esimaion algorihm (Gauss-Newon mehod) is proposed o ge a parameric descripion of he original waveform b a Gaussian. Single responses and more han one response (muliple pulses) are invesigaed. Addiionall, he daa cube opens he possibili of considering neighborhood relaions for space-ime filering or segmenaion. Image slices eraced from he daa cube were processed o esimae line segmens for surface feaures. This has a significan advanage: he esimaed lines are no based on single range poins received from a pulse deecor which delivers more or less accurae range values, bu he are based direcl on he inensi disribuion of he backscaered signal. Since man values of he inensi disribuion have conribued o each line segmen, he accurac of locaion can be much beer han he piel dimension of he image slice. 1. INTRODUCTION The auomaic generaion of 3-d models for a descripion of man-made objecs, like buildings, is of grea ineres in phoogrammeric research. In phoogrammer a spaial surface is classicall measured b riangulaion of corresponding image poins from wo or more picures of he surface. The poins are manuall chosen or auomaicall deeced b analzing image srucures. Besides his indirec measuremen using objec characerisics, which depends on naural illuminaion, acive laser scanner ssems allow a direc and illuminaionindependen measuremen of he range. Laser scanners capure he range of 3-d objecs in a fas, conacless and accurae wa. Overviews for laser scanning ssems are given in (Huising & Pereira, 1998; Wehr & Lohr, 1999; Balsavias, 1999). Curren pulsed laser scanner ssems for opographic mapping are based on ime-of-fligh ranging echniques o deermine he range of he illuminaed objec. The ime-of-fligh is measured b he elapsed ime beween he emied and backscaered laser pulses. The signal analsis o deermine he elapsed ime picall operaes wih analogue hreshold deecion (e.g. peak deecion, leading edge deecion, consan fracion deecion). Some ssems capure muliple reflecions, caused b objecs which are smaller han he fooprin locaed in differen ranges. Such ssems usuall capure he firs and he las backscaered laser pulse (Balsavias, 1999). Currenl firs pulse as well as las pulse eploiaion is used for differen applicaions like urban planning or foresr surveing. While firs pulse regisraion is he opimum choice o measure he ouer envelope of pariall penerable objecs (e.g. canop of rees), las pulse regisraion should be chosen o measure nonpenerable surfaces (e.g. ground surface). Beside he firs or las pulse eploiaion he complee waveform in beween migh be of ineres, because i includes he backscaering characerisic of he illuminaed field. Invesigaions on analzing he waveform were done o eplore he vegeaion concerning he bio mass, foliage or densi (e.g. rees, bushes, and ground). NASA has developed a proope of a Laser Vegeaion Imaging Sensor (LVIS) for recording he waveform o deermine he verical densi profiles in foress (Blair e al., 1999). The spaceborne Geoscience Laser Alimeer Ssem (GLAS) deermines disance o he Earh s surface, and a profile of he verical disribuion of clouds and aerosols. In some applicaions (e.g. observaion of climae changes), clouds are objecs of ineres. In ohers, clouds can be considered as obsacles ha limi he visibili of he illuminaed objec. Recen developmens of laser scanner ssems led o ssems ha allow capuring he waveform wih approimael 1GSample/s: RIEGL LMS-Q560, LITEMAPPER 5600, OPTECH ALTM 3100, TOPEYE II. This waveform sampling rae provides a range resoluion of 0.15m. Our eperimenal ssem measures a a sampling rae of 0GSample/s (0.0075m) and resolves fine srucures wih high accurac. For inerpreaion of his backscaered waveform a general undersanding of he phsical principles is necessar. The phsical measuremen process and he influence of he surface on he emied waveform have been discussed in previous papers (e.g., Juzi & Silla, 00; Wagner e al., 004). In his paper we describe invesigaions for a deailed analsis of laser pulses. In Secion differen echniques for measuremen are discussed. The eperimenal ssem for a fas recording of signals is described in Secion 3. The performed eperimens are eplained in Secion 4 and he obained measuremens are depiced in Secion 5 and 6. We finish wih a discussion abou he received surface feaures.

2 . MEASUREMENT TECHNIQUES 3.1. Receiver uni:.1 Laser scanning ssems Depending on he applicaion, laser scanning ssems can be designed in differen was (Jelalian, 199; Kamermann, 1993). The ma differ in echniques concerning e.g. he modulaion, deecion, or measuremen. Concerning he modulaion echniques laser ssems can be spli beween coninuous wave (cw) laser and pulsed laser. For applicaions in remoe sensing he pulsed laser wih he higher power densi compared o cw laser is of advanage, because i allows operaing a long ranges. In his work, we focus on pulsed laser ssems. Deecion echniques can be spli beween coheren deecion and direc deecion. Coheren deecion considers he wave fron of he received signal compared o a reference signal emied from a cw laser. In direc deecion laser ssems he received opical energ is focused ono a phoosensiive elemen ha generaes a value ha depends on he opical power. Measuremen echniques for range deerminaion can be disinguished b he eploied signal properies like phase, ampliude, frequenc, polarizaion, ime, or a combinaion of hem. We are ineresed in measuring and analzing he received pulse form, i.e. he dependence of he inensi over ime. The classical measuremen echnique for direc deecion operaes wih a phoodiode. The phoodiode generaes an elecrical signal (volage or curren) ha is direcl proporional o he opical power of he inciden ligh (muli phoon). For a deailed analsis of he waveform a digiizing receiver uni is essenial. Analzing he waveform of he emied shor duraion laser pulse wih onl a few nanoseconds requires a receiver uni processing he daa wih a bandwidh of some GHz and an appropriae sampling rae. Increasing bandwidh resuls in decreasing sensiivi of he phoodiode which can be compensaed b increasing power of he emiing laser source. 3. EXPERIMENTAL SETUP An eperimenal seup was buil up for eploring he capabiliies o recognize urban objecs using a laser ssem. For he main invesigaions of he influence of differen objec properies on he waveform, a pulsed laser ssem wih muli phoon deecion is used. The measuremens were carried ou b an eperimenal seup consising of a laser ssem and a capured scene wih pical urban objecs and maerials. The receiver uni o capure he waveform base on an opical o elecrical converer. This converer conain a InGaAs deecor sensiive a he wavelengh of nm. We used a 50MHz receiver for measuring he backscaered waveform, and sampled he daa wih 0GSample/s Moion conrol uni: For he -d scanning process a fiber arra is used for azimuh scan (63 raser seps) and a moving mirror for elevaion scan (30 raser seps). The field of view is ±15 degrees for horizonal and verical direcion. Targes Depending on he applicaion differen properies of urban surfaces can be sensed b a laser ssem (Juzi & Silla, 003b). According o he size of he focused surface geomer in relaion o he beam (fooprin d and wavelengh λ) we divide surface srucures ino macro, meso, and micro srucures (Figure 1). Macro Srucure: We inerpre macro srucures as srucures which are much more eended han he beam fooprin d. Laser range measuremens aken b a scan (which is for aerial surve picall spaced greaer han he spaial beam widh) allow a reconsrucion of large objec srucures like differen roof shapes (e.g. fla roof, gabled roof, hip roof, ec.). Meso Srucure: We inerpre meso srucures as srucures which are much less eended han he beam fooprin d and much greaer han he wavelengh λ. Differen elevaed objec surfaces wihin he beam corridor lead o a miure of differen range values. This ma be caused b small elevaed objecs (e.g. a chimne), a slaned plane, large roughness (e.g. grass, small bushes), or vegeaion (e.g. branches, leaves). Micro Srucure: We inerpre micro srucures as srucures which are less eended han he wavelengh λ. Depending on he roughness he irradiance is more or less refleced. A single measuremen of he backscaered beam inensi (ampliude) gives informaion abou he roughness of he surface and he maerial. Small roughness resuls in specular reflecance and large roughness in diffuse reflecance. Macro srucure > d d > Meso srucure >> λ λ > Micro srucure 3.1 Laser ssem The laser ssem has hree main componens: an emier uni, a receiver uni, and a moion conrol uni Emier uni: We use a shor duraion laser pulse ssem wih a high repeiion rae (4 khz). The pulsed Erbium fiber laser operaes a a wavelengh of 1.55 µm. The average power of he laser is up o 10 kw and pulse duraion is 5 ns (FWHM). differen roof shapes differen elevaed small roughness of he objecs surface and maerial Figure 1. Differen levels of deails sensed b a laser ssem

3 Tes scene 4. EXPERIMENTS According o he focused scale of srucure relevan waveform feaures has o be eamined. A es scene was capured which consis of pical man-made and naural objecs (Figure ). The objecs of ineres for our eperimens are, e.g. buildings, srees, cars, parking slos, rees and meadow. Mos objecs are parl occluded and he maerials have differen backscaering characerisics. Scan The es scene was illuminaed b a pulsed laser wih a wavelengh of appro µm and a beam divergence of appro. 1 mrad. The complee waveform of he backscaered pulse was recorded wih a range resoluion of 8 mm per value. The scene was scanned b 30 elevaion seps in direcion (±15 degrees) and 63 azimuh seps in direcion (±15 degrees). The laser ssem was posiioned 15 m above he ground. Gauss-Newon mehod (Harle & Zisserman, 000) wih ieraive parameer esimaion is used. The esimaed parameers for waveform feaures are he averaged ime value τ (macro srucure), sandard deviaion σ (meso srucure) and maimum ampliude a (micro srucure): a ( τ ) w ( τ ) = ep( ) (1) πσ σ To sar he ieraion, we use he acual parameer values (ime a pulse maimum, widh of signal a half pulse heigh, and pulse maimum) of he original waveform. In Figure 3 an eample for a pical measured waveform is shown b a solid line. The overlaid doed line shows he waveform derived from hese esimaed parameers; we refer o his as he esimaed waveform. 4.3 Daa A 3d daa se was capured b scanning he es scene in and direcion and he waveform over he ime. Inerpreing each inensi value of a waveform as single disance values, hen he measured pulses, which are included in he daa se, can be seen as a spaial disribuion. Assuming Caresian coordinaes he spaial disribuion ma be seen as a daa cube filled up wih inensi values for each (,,) coordinae, where he inensi values depend on he backscaering characerisic of he illuminaed surface. This daa cube can be processed in differen was. Figure 3. Pulse form represenaion: original waveform (solid line) overlaid b he esimaed waveform (doed line) 5.1 Pulse analsis for macro srucures For measuring macro srucures wih eended surfaces we process he daa cube over he ime and assume ha single pulses are received, i.e. we neglec he boundaries (disconinuiies) wihin he fooprin. From he esimaed waveform he averaged ime value τ is used o eploi he emporal form of he received pulses for macro srucures. The averaged range value r can hen easil be deermined wih Figure. Tes scene wih differen urban objecs. 5. 1D ANALYSIS For gaining surface characerisics each waveform of he cube is analzed wihou neighborhood relaions. To characerize he surface he pulses of he waveform have o be eraced. For pulse deecion a noise dependen hreshold is esimaed o separae a single pulse from he background noise. Therefore he background noise is esimaed, and if he inensi of he waveform is above 3σ n of he noise sandard deviaion for he duraion of a leas 5 ns (FWHM of he pulse), hen he waveform will be acceped for furher processing. Tpical surface feaures which we wish o erac from a waveform are disance, roughness, and reflecance. The corresponding waveform feaures of his surface feaures are: ime, widh and ampliude. Because of he srong flucuaions of he waveform, eracing he relevan feaures of he waveform can be difficul. Therefore, he recorded waveform is approimaed b a Gaussian o ge a parameric descripion. To esimae he relevan waveform feaures of he srucures he τc r = () where c is he speed of ligh. 5. Pulse analsis for meso srucures Differen elevaed objec surfaces wihin he beam corridor lead o a miure of differen range values. A plane which is slaned in relaion o he viewing direcion shows differen range values wihin he fooprin. This range inerval which is given b he size of he fooprin and he orienaion of he plane leads o a emporal spread of he pulse (Figure 4b). A deformaion of he pulse form can also be caused b perpendicularl oriened plane surfaces shifed b a small sep in viewing direcion (Figure 4c). A large sep leads o wo separae pulses (Figure 4d). Several surfaces wih differen range wihin he beam resul in muliple pulses. From he esimaed waveform he sandard deviaion σ is used o eploi he emporal form of he received pulses for meso srucures. The increased pulse widh of meso srucures indicaes large roughness of vegeaion or an uneven, slaned or small sepped surface.

4 a b c d Figure 4. Surface and pulse waveform. a) fla surface, b) slaned surface, c) small sep, d) large sep Pulse analsis for micro srucures Depending on he maerial and he surface roughness (micro srucure) objecs show differen reflecance properies. The amoun of backscaered phoons varies wih hese properies. From he esimaed waveform he maimum ampliude a is used o eploi he emporal form of he received pulses for micro srucures. The received maimal ampliude or pulse power can be used o discriminae differen objecs or maerials. Muliple pulses Boh firs pulse and las pulse deecion are used for phoogrammeric applicaions which allow considering or neglecing he presence of vegeaion. Generall, vegeaion leads o a various number of pulse reflecions depending on he densi and srucure. These various number of pulses (we call hem muliple pulses) can be considered b analzing he complee waveform. This migh be of ineres for discriminaing differen pes of vegeaion. Muliple reflecions can also be observed for urban srucures which are smaller han he fooprin (e.g. power lines). Le us invesigae a single signal wih muliple pulses. Therefore an eample of a signal profile consising of muliple pulses is seleced from he measured daa se and i is depiced in Figure 5a. The waveform parameers for each deeced pulse of his signal profile are esimaed and described in Table 1. In conras o he original waveform he esimaed waveform is shown in Figure 5b Figure 5. Waveform esimaion. a) original waveform, b) esimaed waveform B comparing he averaged ime values in Table 1, we can see, ha he disance beween he 1 s and nd pulse is abou 10 m and beween he 3 rd and 4 h pulse abou.5 m. The 3 rd pulse shows he highes maimum ampliude and he sandard deviaion of he 1 s and nd pulse is slighl lower han he 3 rd and 4 h pulse. The quesion arises: can his be helpful o classif he parl illuminaed surfaces onl b inerpreing waveform feaures? I seems ha he characerisics of he waveform feaures are no srong enough o accomplish his. Inside he beam corridor, several parl illuminaed surfaces of various sizes and maerials cause ambiguous waveform feaures. For inerpreing each pulse of a signal profile which includes muliple pulses furher informaion is needed. Onl b considering neighborhood relaions an inerpreaion migh be possible. Pulse Objec Maimum ampliude a [mv] Sandard deviaion σ [m] Averaged ime value r m [m] nd s Vegeaion h rd Building Table 1. Esimaed parameers of waveform feaures To visualize he neighborhood relaions, he measured daa cube (,,) was sliced vericall in - planes. Figure 6 shows a se of image slices (- planes). The nd slice from he lef in Figure 6 shows vegeaion in he cener (near range) and building srucures on he righ side (far range). The gre values correspond o he inensi of he signal. The inensi values along he marked solid line are he inensi values of he waveform described in he paragraphs above. B considering he neighborhood relaions of he waveform he inerpreaion and classificaion of each single pulse is now possible. We can sa ha he crown diameer of he ree is a leas he disance beween he 1 s and nd pulse (10 m). Furhermore we can sa ha he 3 rd and 4 h pulse resul from a building edge. If he scene cone (e.g. vegeaion, buildings) is known, hen an inerpreaion of he pulse feaures ges much easier. Beside he inerpreaion, he processing of muliple reflecions caused b vegeaion can be improved b readjusmen and adapaion of he predeermined hreshold o deec vegeaion srucures wih low reflecance. I has o be remarked ha his kind of slice image in Figure 6 can no give a real side looking view. We sill have o consider ha he daa cube includes.5d and no real 3d informaion. Depending on he area size in relaion o he beam fooprin i is possible, ha he complee pulse inensi is backscaered from he firs illuminaed surface in propagaion direcion and he following surfaces give onl poor or none reflecions. For insance, a ree wih dense foliage ma reurn onl a single reflecion response per laser pulse illuminaion. To give an eample, he scene was scanned wih 0160 scan poins of which 9014 scan poins include a leas a single reflecion response. The number of received reflecions per scan poin in percen wih a leas a single reflecion response (Table ) shows ha 0% of muliple reflecions are received. I was quie rare o receive 4 and more pulses (4%). Number of reflecions and more % Table. Number of received reflecion responses

5 X Figure 6. Verical image slices wih ground, vegeaion and building srucures 4 6. D ANALYSIS In his secion we will presen a wa o analze he daa cube for line segmens. Unil now processing laser scanner daa in form of poin clouds depends on he srengh and weakness of he pulse deecion algorihm. However he received range values of he algorihm in form of poins have hen been used o obain line segmens or planes. Insead of he radiional processing of laser scanner daa in erms of poin clouds, disribued inensi values are now direcl considered. This mehod has he advanage of omiing he pulse deecion algorihm b direcl processing disribued inensi values of he daa cube for higher level segmens in form of lines or planes. A simple wa o analze he daa cube is o slice he complee cube in, and direcion o obain image slices from differen poins of view. These image slices can hen be used for visualizaion as shown in Figure 6 or for processing, as will be shown in his secion. The main goal of his processing here is no o increase he accurac, bu o obain line segmens which give a descripion of he surface on a higher level han poin clouds. Bu before we sar o process he image slices we have o undersand wha we receive if we slice he daa cube. Therefore we suppose a scene wih a plane surface perpendicular o he direcion of he beam propagaion. This surface is scanned in one direcion and he waveform of each scan poin is recorded. For each recorded measuremen we ge a 1d disribuion of inensi values wih respec o ime. B aligning neighboring (in -direcion) measuremens side b side, we ge a - plane of Figure 7. Analzing he daa cube a) verical image slices, b) corresponding line segmens

6 inensi disribuions wih a pronounced maimal line parallel o he -ais. The ime value of his line corresponds o he disance of he surface from he deecor. In our invesigaions, for eample, we slice he daa cube vericall in direcion and receive 63 image slices (- planes). Each single image consising of inensi values induced b he scaering effecs of he surface roughness can now be processed b a line deecor o esimae plane objec surfaces. For sraigh line deecion he Sandard Hough Transform (SHT) (Ballard, 1981) can be used. To appl his algorihm, he region of ineres (ROI), covering he areas where surfaces are epeced, has o be eraced. This can be done b calculaing he inensi disribuion of all inensi values of an image slice and seing all inensi values below σ of he image sandard deviaion o zero. The remaining inensi values above zero (ROI) are furher processed wih he deecor o ge a line descripion of illuminaed plane surfaces. To feed he accumulaor space of he SHT each nonzero piel is ransferred and weighed wih is inensi value. In Hough space we obain superimposed sinusoids wih local peaks. Peak values in his space represen poenial lines in he image slice and have o be eraced and ransformed back o he image space. The lengh of he sraigh line segmen is deermined b an overlap crierion of he line wih he ROI (nonzero piels). Line segmens shorer han a specified value are discarded. Since man values of he inensi disribuion have conribued o each sraigh line segmen, he accurac of locaion can be much beer han he piel dimension of he image slice. An eample of an image slice (Figure 7a) and he esimaed sraigh lines (Figure 7b) are depiced. Pars of he ground and he main building s facade and roof, are esimaed b lines. Processing he measured daa cube ields line descripions for each slice (Figure 8). The ground (lef side and foreground) and he main building (righ side) are deeced, bu also some ree runks. If he raw daa cube is processed, hen he deeced line segmens can easil be ransformed from spherical ino caresian coordinaes. Wheher he inverse procedure of ransforming he daa cube firs and performing he line deecion in caresian coordinaes aferwards ields beer resuls remains o be esed. Furhermore i has o be considered ha onl if he illuminaed surface is larger eended han he lengh of he inensi disribuion or a leas larger eended han he pulse widh, hen he eraced line segmen can have he same orienaion as he illuminaed surface plane. Figure 8. Esimaed line segmens of he invesigaed es scene 7. CONCLUSION In his paper we eplored he capabiliies of analzing he shape of he waveform for surface feaures in form of disance, roughness, and reflecance for each backscaered pulse. Because of he srong flucuaions of he waveform, an ieraive esimaion algorihm is proposed o ge a parameric descripion of he original waveform b a Gaussian. Beside he invesigaions on single responses he eraced parameers of more han one response (muliple pulses) do no lead o significan surface feaures. Onl b considering neighborhood relaions (e.g. maerial characerisic) an inerpreaion becomes possible. Space-ime analsis was inroduced o invesigae he measured daa volume (space-ime cube). The image slices eraced from he daa cube were processed o esimae line segmens for surface feaures. This has a significan advanage: he esimaed lines are no based on single range poins received from a pulse deecor which delivers more or less accurae range values, bu he are based direcl on he inensi disribuion of he backscaered signal. Since man values of he inensi disribuion have conribued o each line segmen, he accurac of locaion can be much beer han he piel dimension of he image slice. Furher work will also deal wih he segmenaion of plane segmens in he 3d daa cube. REFERENCES Balsavias EP (1999) Airborne laser scanning: eising ssems and firms and oher resources. ISPRS Journal of Phoogrammer & Remoe Sensing 54: Blair JB, Rabine DL, Hofon MA (1999) The Laser Vegeaion Imaging Sensor (LVIS): A medium-aliude, digiizaion-onl, airborne laser alimeer for mapping vegeaion and opograph. ISPRS Journal of Phoogrammer & Remoe Sensing 56: Ballard DH (1981) Generalizing he Hough Transform o Deec Arbirar Shapes. Paern Recogniion, Vol. 13, No., pp Harle R, Zisserman A, (000) Muliple View Geomer in Compuer Vision. Proc. Cambridge Universi Press, Cambridge. Huising EJ, Gomes Pereira LM (1998) Errors and accurac esimes of laser daa acquired b various laser scanning ssems for opographic applicaions. ISPRS Journal of Phoogrammer & Remoe Sensing 53: Jelalian AW (199) Laser Radar ssems. Norwood, MA, Boson: Arech House Juzi B, Eberle B, Silla U (00) Esimaion and measuremen of backscaered signals from pulsed laser radar. In: Serpico SB (ed) (003) Image and Signal Processing for Remoe Sensing VIII, SPIE Proc. Vol. 4885: Juzi B, Silla U (003) Laser pulse analsis for reconsrucion and classificaion of urban objecs. In: Ebner H, Heipke C, Maer H, Pakzad K (eds) Phoogrammeric Image Analsis PIA 03. Inernaional Archives of Phoogrammer and Remoe Sensing. Vol. 34, Par 3/W8, Kamermann GW (1993) Laser Radar. In: Fo CS (ed) Acive Elecro- Opical Ssems, The Infrared & Elecro-Opical Ssems Handbook. Michigan: SPIE Opical Engineering Press Wagner W, Ullrich A, Melzer T, Briese C, Kraus K (004) From singlepulse o full-waveform airborne laser scanners: Poenial and pracical challenges. In: Alan MO (ed) Inernaional Archives of Phoogrammer and Remoe Sensing. Vol 35, Par B3, Wehr A, Lohr U (1999) Airborne laser scanning an inroducion and overview. ISPRS Journal of Phoogrammer & Remoe Sensing 54: 68-8.