• No results found

Traffic Signs Detection / Recognition - Proposal and Preliminary Results

N/A
N/A
Info
Downloaden
Protected

Academic year: 2021

Share "Traffic Signs Detection / Recognition - Proposal and Preliminary Results"

Copied!
38
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Download het nu ( 38 pagina )

Hele tekst

(1)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Traffic Signs Detection / Recognition

-Proposal and Preliminary Results

(2)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Overview

Single-view

Segmentation - extremely fast bounding box selection process with FN → 0.

Traffic signs are designed to be well distinguishable from background ⇒ have distinctive colors and shapes.

Detection - AdaBoost classifier of bounding boxes. Recognition - SVM.

(3)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Overview

Single-view

Segmentation - extremely fast bounding box selection process with FN → 0.

Traffic signs are designed to be well distinguishable from background ⇒ have distinctive colors and shapes.

Detection - AdaBoost classifier of bounding boxes. Recognition - SVM.

(4)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(5)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(6)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(7)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(8)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(9)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Thresholding

Segmentation is often detection of connected components of a thresholded image (T = [0.5, 0.2, −0.4, 1.0]>)

Original Thresholded Connected Segmented

image image I (T ) components bound. boxes

Threshold in RGB space: T = [t, a, b, c]> Thresholded image I (T) = ( 1 a· IR +b· IG +c· IB ≤t 0 otherwise

(10)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Thresholding

Segmentation is often detection of connected components of a thresholded image (T = [0.5, 0.2, −0.4, 1.0]>)

Original Thresholded Connected Segmented

image image I (T ) components bound. boxes

Threshold in RGB space: T = [t, a, b, c]> Thresholded image I (T) = ( 1 a· IR +b· IG +c· IB ≤t 0 otherwise

(11)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Learning segmentation

One threshold insufficient ⇒ set of thresholds T = {T1, T2, . . . } used.

Different T different errors:

FN(T ) ... False Negative (missed traffic signs) FP(T ) ... False Positive (segmented backgrounds) card(T ) ... Cardinality (size of T )

Learning segmentation is searching for T ⊆ P{T} minimizing an error subject to some constraints. Learning is NP-complete (set covering problem) but tractable even for very large P(T).

(12)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Learning segmentation

One threshold insufficient ⇒ set of thresholds T = {T1, T2, . . . } used.

Different T different errors:

FN(T ) ... False Negative (missed traffic signs) FP(T ) ... False Positive (segmented backgrounds) card(T ) ... Cardinality (size of T )

Learning segmentation is searching for T ⊆ P{T} minimizing an error subject to some constraints. Learning is NP-complete (set covering problem) but tractable even for very large P(T).

(13)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Learning segmentation

One threshold insufficient ⇒ set of thresholds T = {T1, T2, . . . } used.

Different T different errors:

FN(T ) ... False Negative (missed traffic signs) FP(T ) ... False Positive (segmented backgrounds) card(T ) ... Cardinality (size of T )

Learning segmentation is searching for T ⊆ P{T} minimizing an error subject to some constraints. Learning is NP-complete (set covering problem) but tractable even for very large P(T).

(14)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Experiments - Segmentation

Neyman-Pearson ( = 5%, card(T) = 104 ⇒ card(P(T)) = 2104 )

Boolean linear programming optimization time less 2 sec. card(T ) ≈ 30 thresholds.

FP ≈ 103 bounding boxes per 1236 × 1628 image.

(15)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Feature normalization

Haar features

Feature normalization (RGB, HSL, HSV, gray) Feature selection (greedy construction by AdaBoost)

(16)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Feature normalization

Haar features

Feature normalization (RGB, HSL, HSV, gray) Feature selection (greedy construction by AdaBoost)

(17)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

AdaBoost

AdaBoost results: 154 features selected from 5 · 104features. 77 minutes learning on 2400 training samples.

(18)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

AdaBoost

AdaBoost results: 154 features selected from 5 · 104features. 77 minutes learning on 2400 training samples. 0 10 20 30 40 50 0 10 20 30 40 50 FP[%] FN[%] RGB

(19)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

(AdaBoost) Cascade

(20)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

(AdaBoost) Cascade

(21)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

(AdaBoost) Cascade

(22)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Multi-class detection by shared cascade [Zehnder et

al. BMVC 2008]

(23)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Multi-class detection by shared cascade [Zehnder et

al. BMVC 2008]

(24)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Multi-class detection by shared cascade [Zehnder et

al. BMVC 2008]

(25)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(26)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(27)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(28)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(29)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(30)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(31)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(32)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(33)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(34)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Extension to 3D - Example

Calibrated view 1: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints Calibrated view 2: Segmentation (FP ≈ 103) Detection (FP ≈ 101) Geometry constraints

(35)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Conclusions

Segmentation based on optimal selection of thresholds in color space given a set of constraints in performance.

Detection aimed to be solved with AdaBoost cascades1.

Shared features/cascades2 and post-optimization of the

classifiers3 are directions of work.

Recognition - SVM is in block start.

1

Viola and Jones, CVPR’01 2

Zehnder et al., BMVC’08, and Torralba et al., CVPR’04 3

(36)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Conclusions

Segmentation based on optimal selection of thresholds in color space given a set of constraints in performance.

Detection aimed to be solved with AdaBoost cascades1.

Shared features/cascades2 and post-optimization of the

classifiers3 are directions of work.

Recognition - SVM is in block start.

1

Viola and Jones, CVPR’01

2

Zehnder et al., BMVC’08, and Torralba et al., CVPR’04 3

(37)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Conclusions

Segmentation based on optimal selection of thresholds in color space given a set of constraints in performance.

Detection aimed to be solved with AdaBoost cascades1.

Shared features/cascades2 and post-optimization of the

classifiers3 are directions of work. Recognition - SVM is in block start.

1

Viola and Jones, CVPR’01 2

Zehnder et al., BMVC’08, and Torralba et al., CVPR’04 3

(38)

Traffic Signs Detection / Recognition -Proposal and Preliminary Results Introduction Segmentation Detection Conclusions

Conclusions

Segmentation based on optimal selection of thresholds in color space given a set of constraints in performance.

Detection aimed to be solved with AdaBoost cascades1.

Shared features/cascades2 and post-optimization of the

classifiers3 are directions of work. Recognition - SVM is in block start.

1

Viola and Jones, CVPR’01 2

Zehnder et al., BMVC’08, and Torralba et al., CVPR’04 3

Referenties

Download het nu ( PDF - 38 pagina - 5.60 MB )

GERELATEERDE DOCUMENTEN

Multimodal Detection and Recognition of Persons with a Static Robot

We discussed a wide range of possible segmentation techniques that were ap- plicable for our scenario and the available sensors, and decided to implement two fundamentally

Contextual Feature Abuse Detection T

A limitation of this research might be that this has not been done enough: distribution of implicit and explicit tweets is still skewed and the creation date bias or author bias

Liturgie in die spanningsveld tussen die Gereformeerde tradisie en Charismatiese vernuwing binne die konteks van die Verenigende Gereformeerde Kerk in Suider Afrika

Alte maklik spreek hulle deur hulle daad van wegbreek die banvloek uit oor die hele gemeenskap van die kerk en probeer hulle om ‘n nuwe gemeente daar te stelAlgemene kommissie Leer

Traffic Signs Detection / Recognition - Proposal and Preliminary Results

Traffic Signs Detection / Recognition - Proposal and Preliminary Results Introduction Segmentation Detection Conclusions Feature normalization Haar features.. Feature

Multi-view Manhole Detection, Recognition and 3D Localisation

• Detection – Classifiers based on histograms of Local Binary Patterns as texture

Multiview Traffic Signs Detection / Recognition

Multiview Traffic Signs Detection / Recognition Introduction Single-view Segmentation Detection and Recognition Multi-view 2D optimization 3D optimization Geometrical

Multi-view traffic sign detection, recognition, and 3D localisation

Meaning of the above used abbreviations is the following colour means method described in Section 3.1, TMSER stands for TMSER(ǫ, ∆) = TMSER(0.1, 0.2) , shape is Section 3.2. FN-BB

Bias in normalization: causes, consequences, detection and remedies

Thus, given a not too small reference emission of hydrogen fluoride, the denomi- nator of the normalization formula is much too small, and a much too high normalization result