Collapsible Cubes: Removing Overhangs from 3D Point Clouds to Build Local Navigable Elevation Maps

Transcription

1 Collapsible Cubes: Removing Overhangs from 3D Point Clouds to Build Local Navigable Elevation Maps Antonio J. Reina, Jorge L. Martínez, Anthony Mandow, Jesús Morales, and Alfonso García-Cerezo Dpto. Ingeniería de Universidad de Málaga (Spain) 1

2 OUTLINE 1. OVERVIEW 2. CBCs DATA STRUCTURES 3. COLLAPSING CUBES METHOD 4. EXPERIMENTAL RESULTS 5. CONCLUSIONS 2

3 1. OVERVIEW 3

4 Overview n Elevation maps: Compact 2½ D terrain surface model n 3D point clouds 3D point cloud of an open door Overhangs produce unreliable maps n Goal Identification and removal of overhangs from point clouds Collapsing cubes instead of a point-based gap search Use of coarse binary cubes (CBCs) data structures Non-navigable elevation map 4

5 2. CBCs DATA STRUCTURES 5

6 Coarse Binary Cubes Data Structures (CBCs) E x, y, z Index Each cube has a unique integer index I à (x,y,z,i) 1D CBCs data structures 1 0 V : Binary Vector L : Occupied Cubes Index List 6

7 3. COLLAPSING CUBES METHOD 7

8 Collapsing Cubes Method n PRINCIPLE: Classify 3D points as ground (including vertical obstacles) and overhangs Sorting index list L implies visiting the lowest occupied cubes in the first place 8

9 Collapsing Cubes Method n Implementation with CBCs data structures Sorted list (L) I 0 I 1 I 2 I k I max CBCs Removed cubes Minimum number of empty cubes If ground Else i x, i y, i z Ground Matrix (M) i z i ymax 0 1 i xmax Removed Cubes (R) I r0 I r1 I r2 I rn I rn+1 9

10 Collapsed Cubes Animation 10

11 4. EXPERIMENTAL RESULTS 11

12 Experimental Setup Departamento Departamento de de Ingeniería Ingeniería de de Sistemas Sistemas yy Automática Automática n UnoLaser 3D Scanner: 30 m range 0.7 m above ground n Elevation Maps Subsampling resolution δ = 0.1m 20x10 (meter) Trees Tunnel QUADRIGA 12

13 Results: Point Classification 1st Outdoor Environment: TREES 13

14 Results: Point Classification 2do Outdoor Environment: TUNNEL 14

15 Results: Computational Times n Comparison between point-based and cube-based Point-based Method variance > threshold 2D grid COMPUTATION TIMES MatLab on a Intel Core i7 gap Method /Scene TREES TUNNEL Cube-based 0.21 s 0.37 s Point-based 0.39 s 0.54 s 15

16 Results: Appication to Build Elevation Maps n Standard Elevation Map 16

17 Results: Appication to Build Elevation Maps n Fuzzy Elevation Map 17

18 Results: Appication to Build Elevation Maps n Standard Elevation Map 18

19 Results: Appication to Build Elevation Maps n Fuzzy Elevation Map 19

20 5. CONCLUSIONS 20

21 Conclusions n Simple processing of leveled 3D point cloud that identifies and removes overhang points Efficient data structures from coarse binary cubes An occupied cube is collapsible when a gap is detected For navigation task this gap depends on the mobile robot height n Improvement in computational times with respect to point-based solution Verified in different outdoor environments n Employed to build reliable standard and fuzzy elevation maps n Work in progress Navigation with the Quadriga mobile robot based on local planned paths from the FEMs 21

22 Departamento Departamento de de Ingeniería Ingeniería de de Sistemas Sistemas yy Automática Automática Thank you! Merci! /ajreina 22

23 Integer Index Cubes E 23

24 Results: Computational Times COMPUTATION TIMES MatLab on a Intel Core i7 Method /Scene TREES TUNNEL Cube-based 0,21 s 0,37 s Point-based 0,39 s 0,54 s Occupied Cubes 1,407 (0.5 m) 2,910 (0.5 m) Scan Points 130, ,641 24