0 *

Efficient Collision Detection for Large-Scale Point-Clouds

Takeru NIWA and Hiroshi MASUDA

Terrestrial laser scanners can capture dense point-clouds from engineering plants. Dense point-clouds are useful for collision detection of objects that move in engineering plants. We propose a real-time collision detection method for large-scale point-clouds. In our method, statuses are classified into collision, no collision, and unknown. The unknown status indicates that points are missing at the object position because of occlusion. Collisions are evaluated on depth-maps, which are generated from point-clouds. Two-resolution of depth-maps are used for improving performance. High-resolution depth maps are used only when objects collide on rough resolution depth maps. Our experimental results show that our method could quickly detect collisions between a large-scale point-cloud and mesh models.

Key words: point- cloud, laser scanner, collision detection

1.

b

t a b

c a 3 t C a b t C CAD

b a

m b a b

a G b CAD b

t ac C Gb B 1 b c t c

b t a b

B c b m a

C t ac

G b B t

a t

a m a C

P 1 c b b B m

c a b

C b b a

C C a C C

i B C ac a

CbG b a C CbGt

CAD m CG a C c B

t b c C Figueiredo1)

Pan2) c b b t Octree t C

t Klein3) c 2 b b t

a t i Radwan4) c

b b t

t C C b b G

C C b t

b a bii c B

5) c G t t

a t t

b c b Na O(N) B G b

c N a O(N) Bc b

t B c 2 b t

b t C

G t a t

C b t c

i b b b

ac c C C a a

b aB c C

P bc b aB B

C aB b ac

G C C t

b b c t

G B c b t a

b t b c b ai

* 27 1 23 27 N N

** × 1-5-1 *** × 1-5-1

Fig.1 Collision detection

(1) t G Cb b b

(2) t a

(3) a ac b

2 c b a C 3 c

t a C 4c t t t

5 c a t

b t

2.

b t b t

a i ac t C b G a

t c a

t t

c a t

t 2 a

t a G b t

t 3 (b) i ac° i

° t c

b a 3 (c)a t 2 b

t c t c c

C t C B

b c 3 (d)ba t b a

P B ac

3(d)a a b 2 b a

i b 3 (e)b a

a b t

c 3 (f, g) a a b

b b a b

a C t 3 (h) ba b t b t a

3. 8 5

8 5

b G c t

C b c !!b 2 a b b

t !,!, ! c ! − ! a

b !,! b t

∆! ∆! t a ! tt

b t a b c

C t C

4 8

c t a

a c P

c t b t

a b b c

b b t a

a b b A B Cab X b t Xt

a b2 t a, b, c,

(a) Reflectance image of point-cloud

(b) Depth map

Fig.4 Generation of depth maps

(a) Point-cloud

(b) Projection on 2D plane

(c) Multi-resolution depth map

(d) Depth values of front and back

(e) Projection of virtual models

(f) Comparison with depths

(g) Relationship of depth values

(h) Collision detection

Fig.3 Overview of collision detection process

Fig.2 Process of collision detection

d, e, f t 5a a b A, B, C bt !!, !!, !! X b , !! c b

a

!! = !(!!!!!!!)(!!!)(!!!) +

!(!!!!!!!)(!!!)(!!!)

ac a a t ! − ! a

c ac C b

P C a b a

t a

b b t

c G i b b a

b t= a t G b

* a t a

C G b c b

t ( ( t b

9 b b c, (79 *9 ( *79

t C a P b

a a t b b

c C

i c ac t

b b t a C C a

c t b a U a

t B b c b

c t C 8 5 .12

c b a

a t P C b b

ac a b b

b b t !! b t !!b t !

b !! !! ! b b a

b c b a b a

!! > !! > ! : ( 8 (a) ) G i b C

a G C b c

P C C

!! ≤ ! ≤ !! : B ( 8 (b) ) b aB b

P C ( ! < !! < !! : ( 8 (c) )

c b aB c C

b a C

b a b

c ac C C b c b a

C B B

c B b c 1 b

B b b t B

b C

c b C

Fig.7 Relationships among depth values

Fig.9 Detection of outliers using neighbor points

Fig.8 Object placed in an occluded region

Fig.5 Projection of mesh onto depth map

(a) Perspective image (b) =- image

Fig.6 Distortion of straight lines on =- plane

Fig.10 Generation of multi-resolution depth maps

a

8 5 936

a ac i

B G b t

c b a G b

B b a C

) b b t t

a a p a C

! c a b )

! = !!"#$ Δ!

∆! c ! c p bb ! b B

B ! c c

t ! = 75° p b b a

b ! C p c

4. 8 5

8 5

t ac t

a B t t

c

i c b t

t a t C a t C t

a t a

b t

b U t C 2 b C t

c 1 b 2 b C

t C c b

t16×16 C c

256 b 1a 8 5

c i C t C t C

C b C a

t C t C

a

a t a i C b a C

b b c C b

B b a t C

a b b t !! !! i

b b 2 b C b t !! !! b b C b

G B

(1) !! > !! > !! : 5

(2) !! < !! < !! : 6

b a C

!! C c a C

b aB b C

b B a C

b b !! C

c B

b a C c t

t b C a C

bC C a b

b 16×16 b a b t Cb

t b b B C C c C

b B C b ac

t t b C

a C B b a b

b t Cb B c a B 5000 b

B t a

t b t a

5.

c 20 b t )68 31 t C 134 b

c - 20 B b b a

(a) No collisions

(b) Collisions

(c) Unknown

(d) Unknown

Fig.12 Results of collision detection

Fig.13 Example of point-clouds

Fig.11 Relationships of depth values on multiple depth map

b t t C t a

c )(*. C

b c *, *

b c t

b B b c G *9 b

( 9 9 b (99 B C

a b Pa b

B c G b

*9 b t b a t C

b t

8 5 .12

t a b

t C t a C

c 16×16 t 1 a C

t C i b a 4×4 t 1a b

t C a a

b t C C

a 3 b t C

b t 3 b

c b B a b C

t t a c t

b t C b

t C c a b t

B a a

b a b b

b t

C B b

t C ac a

b b

( c b

ac b

b a

G

b G b

C b t C a

b G

Table.1 Three cases of depth maps Low Middle High

1 Layer - - 5760 11520

2 Layers 360 720 - 5760 11520

3 Layers 360 720 1440 2880 5760 11520

Table.2 Performance evaluation (frame per second) No Collision Collision Unknown Average

1 Layer 25 34 26 26

2 Layers 578 548 544 562

3 Layers 289 284 287 287

4

a b t b t

c b * 79 b

t C a t b G

t a b

b c mG

B b B

B G

a a C b b

t a c C

c b

a C B

B B C c b

B B ) b

G b a C c a C

C G

Table.3 Performance evaluation (frame per second) Triangles No Collision Collision Unknown Average

4800 245 247 247 246

10800 153 158 153 156

14700 125 126 129 127

19200 102 103 104 103

Table.4 Worst cases of processing time (frame per second) Triangles No Collision Collision Unknown

4800 147 154 138

10800 93 112 94

14700 73 87 90

19200 60 63 62

Table.5 Performance evaluation (frame per second) Cube size No Collision Collision Unknown Average

25cm 254 256 259 256

50cm 245 247 247 246

75cm 221 231 229 227

100cm 213 229 216 222

a b b t

c )- a

t a a

t a b G

b a c a b

a C G

6.

c b t G

t c G

C t t

a B

b 3 b t i t C

b t a

t c b G b

t C t C

b b

t C B

c G b a t

a a C C B

i b b a C

C B

1) M. Figueiredo, J. Oliveira, B. Araújo, J. Pereira: An efficient collision detection algorithm for point cloud models, 20th International conference on Computer Graphics and Vision, 43 (2010) 44

2) J. Pan, I. A. Sucan, S. Chitta, D. Manocha: Real-time collision detection and distance computation on point cloud sensor data, In Robotics and Automation (ICRA), 2013 IEEE International Conference on, (2013) 3593-3599

3) J. Klein, G. Zachmann: Point cloud collision detection, Computer Graphics Forum, 23 (2004) 567-576

4) M. Radwan, S. Ohrhallinger, M. Wimmer: Efficient collision detection while rendering dynamic point clouds, In Proceeding of 2014 Graphics Interface Conference, Canadian Information Processing Society, (2014) 25-33

5) : a (2013) M67

6) H. Masuda, I. Tanaka, M. Enomoto: Reliable surface extraction from point-clouds using scanner-dependent parameters, Computer-Aided Design and Applications, 10(2) (2013) 265-277