using histogram correlation to create consistent laser ... · using histogram correlation to create...
TRANSCRIPT
Universität Bremen
Using Histogram Correlation to Create Consistent Laser Scan Maps
Thomas Röfer
Bremen Institute of Safe SystemsCenter for Computing Technology (TZI)
Universität Bremen
Bremen Institute of Safe Systems 2
Universität Bremen
Motivation
without scan matching with scan matching
Bremen Institute of Safe Systems 3
Universität Bremen
Contents
� Scan Matching Approach� Histogram Matching� Uniqueness & Contrast
� Map Building & Self-Localization� Map Representation� Construction of the Map� Distribution of Errors� Examples
� Conclusion & Outlook
Bremen Institute of Safe Systems 4
Universität Bremen
Basic Idea
a
ba
b
Bremen Institute of Safe Systems 5
Universität Bremen
Approach
rotationalshift
rotationalshift
1st translationalshift
1st translationalshift
2nd translationalshift
2nd translationalshift
coarse shift(1/10 resolution)
coarse shift(1/10 resolution)
fine shift(0,25° or 1 cm)
fine shift(0,25° or 1 cm)
projection filterprojection filter
linesegmentation
linesegmentation
generationof histograms
generationof histograms
histogramcorrelation
histogramcorrelation
Bremen Institute of Safe Systems 6
Universität Bremen
Rotational Shift
a b
0°
-60°
30°
90°
30°-60°
90°
0°-90° 90°
Σ lengtha b
Bremen Institute of Safe Systems 7
Universität Bremen
1st Translational Shift
a b
0°-90° 90°
1st main directionΣ length
ab
shift
Σ length
a
b
Bremen Institute of Safe Systems 8
Universität Bremen
2nd Translational Shift
0°-90° 90°
a
b
shift
Σ length
a b
2nd main direction±20°
1st main directionΣ length
a
b
Bremen Institute of Safe Systems 9
Universität Bremen
Uniqueness & Contrast
similarity
offset
uniqueness leftcontrast
rightcontrast
Bremen Institute of Safe Systems 10
Universität Bremen
scan 1scan 1
Map Representation
scan 2scan 2
scan 3scan 3
.
.
.
� distance data� pose
� frame of reference (age)
� 1st main direction� 2st main direction
� left/right contrast� rotation� 1st main direction
� 2st main direction
� distance data� pose
� frame of reference (age)
� 1st main direction� 2st main direction
� left/right contrast� rotation� 1st main direction
� 2st main direction
Bremen Institute of Safe Systems 11
Universität Bremen
Map Building & Self-Localization
selection ofreference scan
selection ofreference scan
determination of shift
determination of shift
priorities:• old• large overlap• close
priorities:• old• large overlap• close
insert predecessorinto map
insert predecessorinto map -
position correction& error distribution
position correction& error distribution
+
distributerotational error
distributerotational error
distributetranslational error
distributetranslational error
Bremen Institute of Safe Systems 12
Universität Bremen
Distributing the Errors
correction of translationcorrection of rotation
Bremen Institute of Safe Systems 13
Universität Bremen
Frame of Reference
17
18
19
20
2122 23 24
25
26
27
2829
30
17
18
18
18
1818 18 18
18
18
18
181818
adjusting the frames of reference
18
18
18
1818 18 18
18
18
18
18181817
1615
14
13
1219
20
2122
23
cycle in a cycle
Bremen Institute of Safe Systems 14
Universität Bremen
Example 1 – MZH 3th Floor
50 m
Bremen Institute of Safe Systems 15
Universität Bremen
Example 2 – MZH 2nd Floor
50 m� Scans� recorded: 9403
� processed: 4208� stored: 122
Bremen Institute of Safe Systems 16
Universität Bremen
Example 3 – University Campus
MZH
NW2 FZB
IW
Bremen Institute of Safe Systems 17
Universität Bremen
Example 3 – Laser Scan Map
400 m
Bremen Institute of Safe Systems 18
Universität Bremen
Conclusion & Outlook
� Scan Matching with Histograms� Projection Filter
� Line Segmentation� Correlation of Histograms with Different Resolutions
� Map Building� In Real-Time (at 84 cm/s)
� Automatic Selection of the Scans Required
� Distribution of Errors
� Outlook� Geometric Map Representation
� Tests in Populated Environments� Probabilistic Approaches