Road Sign Recognition Project...Matlab Implementing In Webots Implementing For Real E-puck Results...
Transcript of Road Sign Recognition Project...Matlab Implementing In Webots Implementing For Real E-puck Results...
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Road Sign Recognition Project
EL BERRIA Nabil
BERNARD Hippolyte
MOURADIAN Alex
EPFL
Bachelor 6 - 2016
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Outline
1 Introduction
2 ExperimentAnalysis Of Road Signs In MatlabImplementing In WebotsImplementing For Real E-puck
3 Results
4 Conclusion
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Goal
Implementation of a C code allowing the e-puck to read asimple road sign with its camera and analyze the picture onboard in order to escape the maze.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Sensors
• IR proximity sensors : to avoid obstacles and stop in frontof the walls
• Camera : to read and analyze the taken image in order totake a decision
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
FFT For Horizontal Stripes
The criteria for selecting the appropriate direction will dependon the values of the FFT magnitude peaks of the vectorscorresponding to a summation of the pixel through the rows,and the columns.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
FFT Sum Of Rows
No variation in the magnitude of the fft.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
FFT Sum Of Colums
Multiple magnitude peaks : noticeable pixel variance in thevertical direction.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Advantage Of Working In Webots
Figure : modelisation of the e-puck and the make in Webots
• The robot and it’s movement are already implemented itallows us to focus only on the road sign recognition part.
• Since we’re simulating on a computer, there are noconstraints on the resources available.
• The debugging is easier, we can simulate thecomportement of the robot much faster.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
E-puck Movement
In order to make the robot advance while avoiding theobstacles of the maze, we implement a function that computeswheels speeds using Braintenberg coefficients. Thesecoefficients were calibrated in order to reach high speeds andincrease performance in the maze.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
E-puck Movement Code
This function is ran indefinitely, until the e-puck is presentedwith a road sign in front of it.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Rotating The E-puckWithout Odometry
• The e-puck microchip has a fixed number of operationsper second.
• We use a wait function where a simple operation(substraction) is executed
• We calibrate the number of processor operations neededto execute a full turn of the e-puck at a speed of 500 :
• For a full turn:
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Stop Criteria And SetupThe criteria on which the e-puck stops in front of a sign isbased on it’s front sensors data. If they reach a certainthreshold value, the robot stops and prepares for taking apicture.
Once the robot is immobile, leds are lit up to increase theluminosity of the image taken.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Black Sign RecognitionThe picture is taken at a format of 40x40 pixels. But becauseof memory limitations, and because Fourier transforms usingthe e-puck library can be done for vectors with 512 values atmost, we isolate two stripes of the picture with 256 valueseach. These stripes correspond to a few vertical and horizontallines of the picture.
Mean pixel value of stripes is computed and if it’s inferior tothreshold value, the sign is black and the robot has to turnback.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Other Road Signs Recognition
A filter is used to increase the picture?s quality. Using themean pixel value we assign black (0) and white (255) to all ofthe pixel values:
We compute the fourier transforms magnitudes for both stripes.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Other Road Signs Recognition
If the black patterns are horizontal, the sum of the fftmagnitude values of the vertical stripe should be higher thanthe fft magnitude values of the horizontal stripe, and viceversa. We base our direction decision based on this criteria.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Percentage Of Succes
Success percentage for different light conditions (40 tests foreach road sign). (a) is the black sign, (b) and (c) arerespectively the horizontal and vertical and the (d) and (e) thehorizontal with noise and vertical with noise.
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Performance In The Maze
Presentation Video
We got an average time of 36.4 [s]
Road SignRecognition
Project
EL BERRIANabil
BERNARDHippolyte
MOURADIANAlex
Introduction
Experiment
Analysis OfRoad Signs InMatlab
Implementing InWebots
ImplementingFor Real E-puck
Results
Conclusion
Conclusion
• Finding a criteria for recognition in matlab
• Implementing using Webots
• Implementing for the real e-puck
• Difficulties
• What we learned?