AR 1,a) 1 1 AR AR AR 5 1deg AR,,, AR Markers Enabling Accurate Pose Estimation even in Frontal bservation Tanaka Hideuki 1,a) Sumi Yasushi 1 Matsumoto Yoshio 1 Abstract: AR markers are useful tools for eas AR, but have a serious problem that the accurac of pose estimation gets worse especiall in frontal observation. We solved the problem b developing novel AR markers based on a new principle. The proposed markers generate moiré patterns b lenticular lenses or a microlens arra, which seem to move according to the visual-line angles. We realized high-accurac and stable pose estimation b image processing of the patterns. Kewords: AR marker, pose estimation, lenticular lens, microlens arra 1. AR Augmented Realit ARToolKit[1] 1 AIST, Tsukuba, Ibaraki, 35 858, Japan a) hideuki-tanaka@aist.go.jp [] AR AR AR AR AR c 1 Information Processing Societ of Japan 1
AR. AR AR Lenti- Mark[3] 1 LentiMark AR ARToolKitPlus 1 3 ArraMark ArraMark 5). ID ArraMark 9 1 Lens area Reference points () ArraMark prototpe 5 1 9 Design of ArraMark. Paste Lenticular lens Stripe pattern 3. 7 1 Pattern variation of LentiMark LentiMark and the structure of variable moiré pattern. Microlens arra Stripe pitch is slightl wider than lens pitch Angle change Black part moves Print Crisscross pattern Structure of lens-area of ArraMark. 3 Principle of moiré movement (Reprint from [3]) LentiMark AR- ToolKitPlus LentiMark 1deg 3. AR 7 Microlens (A) Center of lens Crisscross pattern (B) Center lens Relation between lenses and crisscrosses. 3.1 LentiMark AR ArraMark 7) c 1 Information Processing Societ of Japan
3.3 ArraMark 8 9 [] Rectangle detection Resize DF pose estimation Modif attitude Four points detection Sampling points (7 +1) 11 [piel] Calculate visual-line angle Scan Transform 3[piel] Binarize Mask processing <Fine> <Coarse> Localization of crisscross Mask image 5[piel] 1 Pose estimation procedure 8 Movement of crisscross according to seeing angle ( c, c ) k θ v c Integrated image Visual lines Lenses Crisscross pattern Vertical line Viewpoint L Visual line crisscross 9 Principle of movement of image according to visual-line angle. 11 Definition of two-dimensional visual-line angles. ArraMark.1., 1.1 AR ArraMark H CM. 1.3 θ ϕ 11 θ, ϕ θ ϕ 1 1 c c c = ( ) T c = k = c - c = c - c = tan -1 ( / ) Calculation of visual-line angles from crisscross position.. H AM =H CM R mod (1) H AM ArraMark H CM R mod H CM R mod c 1 Information Processing Societ of Japan 3
Pv r Pv c 13,1 13 d L P crisscross from crisscross Calculation of viewpoint 1 P from crisscross P vr from reference points Homogeneous transform H ρ ρ Rotation ais Pose modification..1 Pv c (1) θ ϕ Lv c () Lv c d Pv c d H CM.. Pv r Pv c ρ H ρ. R mod R mod =H 1 ρ. ArraMark (1) (1) H AM H CM ArraMark 5. ArraMark ArraMark 5.1 ArraMark ( 1 1) T USB. CMS ARTCAM-3SS-BW (ARTRAY C.,LTD.) 758 [piel] CPU bit,.8ghz, RAM 8GB) C/C++ pencv.3. 5. 5..1 ArraMark 5.1 8[Hz] 5.. [piel] 5..3 ArraMark ±1[deg] ArraMark 9[piel] 15. 1 1 1 8 Z-ais direction error [deg] Conventional method ArraMark -57 - - -1-8 - 8 1 58 Scale 1/5 Scale 1/5 Rotation angle [deg] 15 Average of result [deg] (Rotation angle: -1~11[deg]) Conv.method ArraMark.81.3 std.dev. 1.51.1.1.7 std.dev. 1.5.9 3.51.1 std.dev..11.13 Comparison of ais direction error between ArraMark and conventional method ArraMark 5.. θ 1[deg] 1[deg] ArraMark 5..5 1 ArraMark 5.. 17 c 1 Information Processing Societ of Japan
Angle error [deg] #1 Coventional method ArraMark Marker size: piel # #1 # #3 # Far Near bservation distance Marker size: 5piel 1 Angle error of perpendicular line of marker according to observation distance ArraMark Angle error [deg] 8 Coventional method ArraMark #1 # #3 # #5 # #7 Dark Bright Brightness #1 # #7 Marker size: 5piel 17 Variation of angle error of perpendicular line of marker according to illumination. Conclusion AR LentiMark ArraMark AR AR CG [1] ARToolKit Homepage: http://www.hitl.washington.edu/artoolkit/ [] Y.Uematsu, H.Saito: Improvement of Accurac for D Marker-Based Tracking Using Particle Filter, inproc. 17th Intl. Conf. Artificial Realit and Teleistence 7, pp.183 189, 7. [3] H. Tanaka, Y. Sumi, Y. Matsumoto: A Visual Marker for Precise Pose Estimation based on Lenticular Lenses, Proc. 1 IEEE International Conference on Robotics and Automation (ICRA1), pp.5-57, 1.5.1-18,St. Paul, MN, USA. [] M. G. Lippmann: Epreuves reversibles donnant la sensation du relief, J. Phs. (Paris) 7, 81 85 (198). c 1 Information Processing Societ of Japan 5