FAST CONTINUOUS 360 DEGREE COLOR 3D LASER SCANNER

FAST CONTINUOUS 360 DEGREE COLOR 3D LASER SCANNER Awu Zhang a, *, Shaoxng Hu b, Yuln Chen a, Hayun Lu b, Fan Yang a, Ja Lu a a Key Laboratory of 3D Informaton Acquston and Applcaton of Mnstry of Educaton, Captal Normal Unversty, Beng 100037, zhangaw163@163.com b School of Mechancal Engneerng & Automaton, Beng Unversty of Aeronautcs and Astronautcs, Beng 100083, husx@buaa.edu.cn WG I/2 - SAR and LDAR Systems KEY WORDS: 3D laser scannng, Data Acquston, Multvew geometry, Calbraton, Texture mappng ABSTRACT: There are many needs for the ablty to fast acqure 3D data from envronmental surroundngs, such as navgaton, mappng, localsaton and robot moblty, fre and polce plannng, urban plannng, but the technology for acqurng dense, wde rangng, accurate 3D data s too expensve to be used wdely. A low-cost, 360 0 contnuous scannng, portable 3D laser scanner s presented. The accuracy of the portable laser scanner was analyzed n detal, and an algorthm of systematc measurement error compensaton was gven. On the other hand, a sequence of mages was captured by a hand-held dgtal camera, and then 2D mages were fast mapped on 3D pont cloud by a method proposed based-on multvew geometry. The expermental results show that the portable laser scanner has hgher measurng accuracy and better data qualty. Its max range of the scanner s 80m, and ts accuracy can acheved 6 mm. The portable laser scanner can measure black obects, t well sut tunnel measurement. The weght of the portable laser scanner s less than 6kg, t can be nstalled on unmanned arcrafts. 1. INTRODUCTION 3D laser scanner s a new actve remote sensng system developed n the last ten years, and t can scan thousands of ponts to a pont cloud of 3D data of the obect and be operated at nght. To a certan extent, 3D laser scanner meets the data demands of some applcatons such as cty plannng, fre and polce plannng, underground mne measurement, tunnel measurement, and so on. 3D laser scanners have a huge applcaton market. Unhapply, up to now, there s stll no homemade 3D laser scanner on Chnese market. The foregn commercal 3D laser scanner has a hgh prce, the terrestral 3D laser scanner s about 1,000,000, and arborne 3D laser scanner s about 15,000,000. In ths case, we developed a low-cost, 360 0 contnuous scannng, portable 3D laser scanner. The portable laser scanner can collect no less than 8000 ponts every second to meet the needs of fast acqurng 3D data from ndoor and outdoor scenes, but t only costs about one ffth of the prce of the foregn commercal 3D laser scanner. In practce, foregn commercal long-range or md-range 3D laser scanner manufactures are not too many. There are sx man manufacturers: Calldus, Leca, Mens, Regl, OpTech and I-Ste. I-Ste does not produce ts own laser sensor and the laser sensor employs Regl s. Calldus, I-Ste, Mens and Regl use the mrror to complete vertcal scannng, whch the horzontal scannng depends on a servo motor rotatng to complete, and so ther laser scanner can scan 360 0. Some laser scanners measure dstances wth the flght tme and some others use a phase-dfference method. By computng the angles the coordnates of the pont n the space are obtan, for example, Leca HDS6000 s phase-based laser scannng. We employ the flght tme method to measure dstances from sensor to arbtrary ponts on the obect surface. As basc scannng devce a tme-of-flght 2D SICK laser scanner (LMS291 or LMS200) s used, and whch s combned wth an addtonal servo drve to reach the thrd dmenson. For our laser scanner desgn, accurate synchronzaton of the laser measurement and the scannng devce s very mportant, and the compensaton of systematc errors s also a key task. Moreover, we also dscuss an approach to map 2D mages from a hand-held camera onto 3D laser ponts. The paper ncludes four man sectons: (1) Desgn of the 360 0 contnuous scannng, portable 3D laser scanner, (2) Error correcton of the 3D laser scanner, (3) Fast mappng the mages onto the pont cloud, (4) Analyss of expermental results. 2. DESIGN OF THE PORTABLE 3D LASER SCANNER 2.1 Structure Desgn The key components of the 3D laser scanner are a 2D SICK laser scanner (SICK LMS291 or SICK LMS200) and a rotatng platform drven by a servo motor. The lne scannng of the 2D SICK laser scanner can form the two modes wth 180 and 100. The samplng resoluton n the modes can be set at 0.25 0, 0.5 0 and 1 0, the max scannng range s 80m, and the scannng accuracy can be up to mm, the scannng baud rate can be set at 9600, 19200, 38400 and 500k. SICK LMS200 or SICK LMS 291 communcates wth the computer by seral nterface, whose maxmum speed can reach 75 tmes/sec, but the nteror bufferng capacty s lmted (max: 812 bytes). If we can not output the scans n tme, the cache wll be refreshed automatcally, so that the part of the data * Correspondng author. 409

The Internatonal Archves of the Photogrammetry, Remote Sensng and Spatal Informaton Scences. Vol. XXXVII. Part B1. Beng 2008 would be lost. General seral nterface, whch s the seral nterface RS232 of the computer, can not meet the needs of the transmttng speed (500k Baud), so t can not transmt the data completely and entrely. We nsert the Quatech Card nto PC wth PCI slots so that the transmsson baud rate can reach 500k. The rotatng platform communcates wth the computer by RS232.The structure of the portable 3D laser scanner s shown n fgure 1. constant accelerated rotaton, ts rotaton angle s computed as follows. 2 ω = t (1) Here, s angle acceleraton, ω s rotaton angle, and t s tme. If the angular velocty s known as v and the accelerated tme of the rotatng platform s known as Δt, the rotaton angle s descrbed as follows. ν ω = 2 t Δ t (2) Fgure 1. The structure of the portable 3D laser scanner If the nterval between scanlnes s Δt v, and then Δt = n Δt v. By testng, n=10, the results are the best. Fgure 3 shows the results of moton compensaton. Fgure 2 s the 3D laser scanner developed by us, ts vertcal scanlne rotates around the z-axs contnuously untl up to 360 0, and the scan angle of the vertcal scanlne can be up to 180 0, and these scanlnes form a closed sphercal scan regon. (a) (b) Fgure 3. The results of moton compensaton. Top: before compensaton; Down: after compensaton Fgure 2. The Portable 3D laser scanner prototype system Top: The Portable 3D laser scanner prototype; Down: The closed sphercal scan regon. 2.2 Moton Compensaton Whle the rotatng platform startng to rotate, before runnng at unform velocty, there s a short-term accelerated process. Supposed the rotatng platform startng s consdered as 3. ERROR CORRECTION If the samplng resoluton vertcal scannng s set at 0.25, then every vertcal scanlne ncludes 401 ponts, t spends 45s for a yawng scan wth 0.25 horzontal resoluton. Supposed the center of the rotaton axs and the center of the mrror wheel of the laser scanner are same, then, 410

The Internatonal Archves of the Photogrammetry, Remote Sensng and Spatal Informaton Scences. Vol. XXXVII. Part B1. Beng 2008 x y z = r = r = r cos β cosα snα sn β cosα α β s the step angle of the rotatng mrror of the 2D laser, s the step rotaton angle, and are respectvely the step number. Actually, the portable 3D laser scanner has system errors: (1) Installaton error l. There s a translatonal offset l between the center of the rotaton axs and the center of the mrror wheel of the laser scanner. (2) Range error Δ ρ, whch results from the obect s surface features, ar humdty, tme-gauges bult-n to the equpment and the reflected energy, etc. (3) Scan angle errorϕ. When the rotatng platform begns to move, t wll spend some tme to reach the constant speed, that results n that the of actual reference drecton and the axs of coordnates can not be are not overlappng, then formula (3) should be modfed as below: (3) poston between the 3D range and 2D mage sensors sacrfces the flexblty of 2D mage capture. In fact, because of occlusons and self occlusons, the methods above descrbed are not sut to the large-scale scenes. We use a hand-held dgtal camera to take the mages from dfferent angles, n dfferent tmes, n dfferent focal length. It s a techncal challenge ntegratng the mages from freely movng cameras wth 3D models or 3D pont clouds. Some related works have done by [Stamos I., 2008, Zhao W., 2005.]. I.Stamos s methods assume the exstence of at least two vanshng ponts n the scene and regster ndvdual 2D mages onto a 3D model. W. Zhao s methods algn a pont cloud computed from the vdeo onto the pont cloud drectly obtaned from a 3D sensor. We use W. Zhao s methods to mappng mages onto pont clouds. (1) Recover mult-vew relatons from an mage sequence by structure and moton. (2) Compute dense depth map usng mult-vew stereo. (3) Determne the camera poses by algnng 3D pont clouds from the camera and the 3D sensor usng ICP (Iteratve Closest Pont). Fgure 4 s a result of texture mappng. xˆ yˆ zˆ = ( ρ + Δρ)cosα cos( β + ϕ) + l cos( β + ϕ) = ( ρ + Δρ)cosα sn( β + ϕ) + l sn( β + ϕ) = ( ρ + Δρ)snα (4) ρ Here,, α β, are known; Δ ρ, l and ϕ are unknown, so the parameters need further calbrated and corrected. In order to calbrate the values of Δ ρ, l and ϕ, we select the tablet calbraton approach. We made 5 tablets, then put two of the tablets at the perpendcular drecton of x-axs (+, -) of the actual reference coordnate system (tablet 1 and tablet 2), another two at the perpendcular drecton of y-axs (+, -) of the actual reference coordnate system (tablet 3 and tablet 4), and the last one at the perpendcular drecton of z-axs (+) (tablet 5). Then, the x-coordnate of the laser pont on tablet 1 can be known accurately, X=L1. In a smlar way, the x-coordnate of the laser pont on tablet 2 s L2; the y-coordnate of the laser pont on tablet 3 s L3; the y-coordnate of the laser pont on tablet 4 s L4; the z-coordnate of the laser pont on tablet 5 s L5. L1 to L5 are vertcal dstances between tablets and the orgn of the actual reference coordnate. Durng calbraton, we moved the tablets to change the values of Ln, so we acqured a set of equaton, and then we performed a compensatng computaton utlzng addtonal parameters. After that, we used sgnfcance test to verfy the parameters sgnfcance, to solve correcton parameters. 4. FAST MAPPING IMAGES ONTO POINT CLOUDS In practce, the scanned data s not contnuous, although contans contnuous colour nformaton. 2D mages mappng on 3D ponts s satsfactory for some applcatons. The tradtonal methods are realzed by rgdly attachng a camera onto the range scanner and thereby fxng the relatve poston and orentaton of the two sensors wth respect to each other [Fr uh C., 2003, Sequera V., 2002, Zhao H.,2003]. Fxng the relatve Fgure 4 3D mage of the gate of the unversty Top: 3D reflectance mage; Down: 3D colour mage. 5. ANALYSIS OF EXPERIMENTAL RESULTS Generally, the qualty and accuracy of the recorded 3D ponts of laser scanners are man parameters, whch affect on the applcaton felds of the laser scanners. Techncal data for our laser scanner s shown as follows. Measurement range 0.5m to 80m Accuracy 6mm Measurement rate up to 8000/sec Laser wavelength near nfrared Vertcal (lne) scannng range 0 0 to180 0 Horzontal (frame) scannng range 0 0 to360 0 Weght 6kg We dd a lot of experments to test the portable 3D laser scanner ncludng data qualty, optmal measurement range, nfluence of surface reflectvty, envronmental condtons. 411

The Internatonal Archves of the Photogrammetry, Remote Sensng and Spatal Informaton Scences. Vol. XXXVII. Part B1. Beng 2008 (1) Test of data qualty We scanned the wall of a buldng usng the portable 3D laser scanner (fgure 5(a)). Seen from the enlarged part, the laser samplng ponts are dstrbuted n order, and there are few ump ponts on the wall plane. It shows the data qualty s very good. Fgure 5(b) s the 3D pont cloud of our researchng room. We can see the door, the wndows, the cabnets, the desks, the chars, the clock on the wall, and the lamps on the floor clearly. The edges of obects are also seen clearly. (2) Test of optmal measurement range The max measurement range s 80m, but the optmal measurement range s from 0.2m to 40m. Whle we captured 3D data of outdoor scenes, we found out that the obects wthn about 40m from the laser scanner to obects are descrbed as detaled as shape nformaton, for example, fgure 6. It s sutable for acqurng 3D data of long galleres and tunnels. We collected 3D data of the prayer-wheel gallery of Jokhang Temple usng the 3D laser scanner to (fgure 7). (a) Plane test (b) Edge test Fgure 5. Test of qualty data Fgure 6. Test of optmal measurement range Fgure 7. The part of the prayer-wheel gallery of Jokhang Temple 412

The Internatonal Archves of the Photogrammetry, Remote Sensng and Spatal Informaton Scences. Vol. XXXVII. Part B1. Beng 2008 (a) (b) (c) Fgure 8. Test of nfluence of surface reflectvty (a) (b) (c) (d) (e) (f) (g) (h) Fgure 9. Test of envronmental condtons (3) Test of nfluence of surface reflectvty Whle we captured 3D data of our researchng room, we found that the data of black chars s entre. So we used the portable 3D laser scanner to scan a black car, the 3D pont cloud data of the black car s very good (Fgure 8(a)). The portable 3D laser scanner can measure black obects, and t s easy appled to measure terrans coal ples. We used t to acqure the 3D data of the coal ples n Tann Huanghua, bult 3D model of the coal ples, gave the profle maps of the coal ples, and compute the volumes of the coal ples, ust lke fgure 8(b)-8(c). (4) Test of envronmental condtons Lasers operate n a very lmted frequency band. Therefore flters can be appled n the recevng unt allowng only ths frequency to reach the recever resp. the camera. If the radaton of the llumnaton source (sunlght, lamps) s strong as compared to the sgnal, enough of ths ambent radaton wll pass the flter and nfluence the accuracy or prevent any measurements at all[boehler, w., 2003]. We employed a glass plate wth the black and whte chessboards to test the nfluence of envronmental condtons. To prevent from the nterference of envronment, we put the glass plate on the playground. We found that dust s present lead to the edge blur (fgure 10(a)). We also found that the angle between the laser beam and the lght lnes of sun s to effect on the scanned data, whle the angle s closer to 0 0, the scanned data s very bad (fgure 10(b)), whle the angle s more than 90 0, there s a lttle negatve nfluence on the scanned data (fgure 10(c)), and the angle less than 90 0 s the most deal (fgure 10(d)). Fgure 10(e)-(h) are the correspondng photos descrbed above. 6. CONCLUSIONS We developed a 360 0 contnuous scannng, portable 3D laser scanner, whch can be appled to many felds, such as cty plannng, ndustral survey, vrtual realty, and dgtal preservaton of cultural hertage. The comparson of ts all Techncal data wth those of foregn 3D laser scanner s shown n Table 1. It shows techncal data of our 3D laser scanner (AX200) reaches these of foregn products of ths knd. In addton, t has the advantages of small volume and low weght about only 6 kg, so t s easy to take, and t only costs about one ffth of the prce of the foregn commercal 3D laser scanner. 413

The Internatonal Archves of the Photogrammetry, Remote Sensng and Spatal Informaton Scences. Vol. XXXVII. Part B1. Beng 2008 Because t can measure black obects, t can used n autonomous tunnel mappng. In the future, we wll develop an autonomous tunnel mappng system based on our 3Dlaser scanner. REFERENCES Boehler, w., 2003. Investgatng Laser Scanner Accuracy. CIPA. Symposum, Turkey. Cheok, G.S., 2002. Calbraton experments of a laser scanner. US Natonal Insttute of Standards and Technology, Report No NISTIR 6922, Sept. Fr uh C., 2003. Constructng 3D Cty Models by Mergng Aeral and Ground Vews. Computer Graphcs and Applcatons, 23(6), pp 52 11. Gelsdorf, F., 2004. A concept for the calbraton of terrestral laser scanners. Proc. of FIG Workng Week, Athens, Greece. Hartley, R., 2003. Multple Vew Geometry n Computer Vson, second edton. Cambrdge Unversty Press. Hu,S., 2006. Regstraton of Multple Laser Scans Based on 3D Contour Features. Proc. of Internatonal Conference on Informaton Vsualsaton (IV06). Hu,S.,2005. Real 3D dgtal method for large-scale cultural hertage stes. Proc. of the Nnth Internatonal Conference on Informaton Vsualsaton, pp. 503~508. Nüchter, A., 2007. 6D SLAM - 3D Mappng Outdoor Envronments. Journal of Feld Robotcs, 24 (8/9), pp. 699 722 Paulo, D., 2006. 3D Reconstructon of Real World Scenes Usng a Low-Cost 3D Range Scanner. Computer-Aded Cvl and Infrastructure Engneerng, 21, pp. 486 497. Sequera, V., 2002. 3D Rrealty Modelng: Photo-realstc 3D Models of Real World Scenes. In Intl. Symposum on 3D Data Processng, Vsualzaton and Transmsson, pp.776 783. Stamos, I., 2008. Integratng Automated Range Regstraton wth Multvew Geometry for the Photorealstc Modelng of Large-Scale Scenes. Internatonal Journal of Computer Vson [Specal Issue on Modelng and Representaton of Large-Scale 3D Scenes]. Wulf, O., 2007. Ground Truth Evaluaton of Large Urban 6D SLAM. Proc. of the IEEE/RSJ Internatonal Conference on Intellgent Robots and Systems. Zhao, W., 2005. Algnment of Contnuous Vdeo onto 3D Pont Clouds. IEEE Transactons on Pattern Analyss and Machne Intellgence, 27(8), pp. 1305 1318. Zhao H., 2003. Reconstructng a Textured CAD Model of an Urban Envronment Usng Vehcleborne Laser Range Scanners and Lne Cameras. Machne Vson and Applcatons, 14(1):, pp. 55 41. Zhang, A.,2004. Extractng 3D Contour Features of Urban Scenes from Ground-Based Laser Range Data. Proc. of IEEE Conference on Informaton Vsualzaton, pp.133-138. Zhang, A., 2004. Constructng 3D realty models of Urban Scenes from Ground-Based Synchronzed Laser and Vsual Data. Proc. of the IASTED Internatonal Conference on Computers, Graphcs and Imagng, pp. 248-253. ACKNOWLEDGEMENTS Ths work s supported n part by Natonal Natural Scence Foundaton of Chna (NSFC 40601081), Beng Scence and Technology Nova proect (2006B57) and Natonal Scence and Technology Support Program (2006BAJ15B01-02) Pollefeys, M., 2004. Vsual Modelng Wth a Hand-Held Camera. Internatonal Journal of Computer Vson, 59(3), pp. 207 232. Type Calldus Cyrax250 I-STE GS 100 ILRIS-3D LMS-Z420 AX100 0 Laser class 1 2 1 2 1 1 2 wavelength (nm) 905 532 904 532 1540 904 532 Measurement range (m) 80 100 450 100 800 200 80 Accuracy (mm) 5( 30m ) 6(50m) 8( 300m ) 6( 100m ) 7( 100m ) 10(100m) 6( 32m ) Measurement rate (pts) 3300 1000 6000 1000 2000 9000 7256 Vertcal scannng range (deg) 360 40 340 360 40 360 360 Vertcal scannng range (deg) 180 40 80 60 40 80 180 Table 1 Comparson wth foregn products n the techncal parameter 414