Automatic vehicle license plate detection using VEDA
-
Upload
rojith-thomas -
Category
Education
-
view
1.636 -
download
9
description
Transcript of Automatic vehicle license plate detection using VEDA
AUTOMATIC VEHICLE IDENTIFICATION USING
VEDA
GUIDED BY; PRESENTED BY; MRS. MERCY MATHEW ROJITH THOMAS ASST. PROFESSOR MTECH-CE CAARMEL ENGG. COLLEGE ROLL NO: 18
INTRODUCTION As number of automobiles grows rapidly, the traffic problems
increase as well, for example, car theft, over speeding and running on the red light.
To avoid these problems, an efficient real time working vehicle identification system is needed.
Most widely accepted technique is License Plate Detection(LPD).
Based on Image processing by capturing license plate using cameras.
Applications:1) crime prevention2) parking and toll fee system3) traffic data collections
BASIC DIAGRAM
Three parts:1) License Plate Detection2) Character Segmentation3) Recognition
EXISTING algorithm
Difficult to process under complex conditions. Kim et al Algorithm: statistical features and templates Zimmermann and Mattas Algorithm: fuzzy logic Sobel Algorithm: vertical edge extraction Canny Algorithm: Vertical edge extraction Abolghashemi Algorithm: low quality input Zhang et al Algorithm: reduce complexity Bai et al Algorithm: stationary and fixed background
PROPOSED algorithm Detection is by extracting vertical edges.
Low quality images are produced by using web camera.
Resolution is of 352 X 258 with 30 fps.
Steps:1) Pre-processing.2) Vertical Edge Detection.3) Plate Extraction.
1. Pre processing Process of generating binarized image from color image. Two steps;
1) Color to gray image inversion(C2G).2) Adaptive Thresholding.
COLOR TO GRAY IMAGE CONVERSION Converting color image into grayscale image.
GRAY IMAGECAPTUARED IMAGE
ADAPTIVE THRESHOLDING Gray image is converted into binarized image. To get good adaptive threshold image , Integral image
technique is used. Earlier technique: Wellner’s Algorithm.
a)Pixel is compared with avg. of neighboring pixels(S). b)Value of S=1/8 of (image). c)If current pixel is T% lower than S, then set to Black. d)Otherwise set to White. e)Value of T=0.15 of (image).
Limitation: Not suitable when samples are not evenly distributed in all directions(Moving System).
INTEGRAL IMAGE FORMULATION Window concept. Image is as matrix with m rows and n columns. Algorithm: Initially, summation of pixel values for every column is
calculated as; sum(i)|j
g(x,y) = input values.sum(i) = all gray value for every column j
through all rows i(i=0,1….m).
1,0 . . . . . . . 1,n
2,0
.
.
.
m,0 m,n
Integral image can be calculate as;
where, IntrgImg(i,j) = integral image for pixel(i,j). Next step is thresholding for each pixel.
1)Calculate intensity summation for each window.2 subtraction and one addition is performed.
i+s/2,j+s/2
i+s/2,j-s/2
i-s/2,j+s/2
i+s/2,j+s/2
Compare value g(i,j) with threshold value t(i,j).
After comparing we get output as;
THRESHOLD IMAGE
2.VERTICAL EDGE EXTRACTION Extracting the data by distinguishing the plate region. Two steps:
a) Unwanted Line Elimination Algorithmb) Vertical Edge Detection Algorithm
UNWANTED LINE ELIMINATION ALGORITHM To avoid long foreground lines and short noise edges
besides LP region(Unwanted Lines) Cases : 1) Horizontal with angle 0⁰(-).
2) Vertical with an angle 90⁰(|).3) Line inclined at an angle 45⁰(/).4) Line inclined at an angle 135⁰(\).
CONCEPT: Black pixel values are the background and White pixel
values are the foreground. A 3X3 mask is used throughout all image pixels from left
to right and from top to bottom Only black pixel values in the image are tested.
b(x,y)
Different cases of converting the centre pixel into foreground
Output as
THRESHOLD IMAGE ULEA OUTPUT
VERTICAL EDGE BASED DETECTION ALGORITHM To find beginning and end of each character Concentrates on intersection of Black-White and White-
Black regions.
A 2X4 mask is used to process the image
Output is as;
Comparing with old edge extraction method
SOBEL METHOD VEDA
3.PLATE EXTRACTION
To extract plate region and characters
Four steps:1) Highlight Desired Details(HDD).2) Candidate Region Extraction(CRE).3) Plate Region Selection(PRS).4) Plate Detection(PD).
HIGHLIGHT DESIRED DETAILS Performs NAND-AND operation for each two
corresponding pixels values taken from ULEA &VEDA. Connecting to vertical edges with black background.
HDDVEDA
hd
NAND AND PROCEDURE
hd is the length between two edges. Computed using test images. Help to remove long foreground lines and noisy edges. Process take place from top to bottom and left to right. After this , plate region exists are highlighted.
VEDA OUTPUT HDD OUTPUT
CANDIDATE REGION EXTRACTION To find exact LP region from the image. Process divide into four steps.COUNT THE DRAWN LINES PER EACH ROW No of horizontal lines in each rows are counted Stored in a matrix variable : lines[a] ;a=0,1……m-1 Time consuming process.DIVIDE THE IMAGE INTO MULTIGROUPS To avoid delay, images convert to multiple groups Stored value in a variable : groups
groups=height/C.C=CRE Constant (10)
COUNT SATISFIED GROUP INDEXES AND BOUNDARIES To eliminate unsatisfied groups which exists in the LP
A threshold value will be considered.Threshold>=1/15 of image height
SELECTING BOUNDARIES OF CANDIDATE REGION More than one region will be present Drawing horizontal line above and below each candidate
region
OUTPUT AFTER CRE
PLATE REGION SELECTION AND DETECTION To extract one correct LP Two steps
1. Selection of LP region2. Making a vote.
SELECTION OF LP REGION Check blackness ratio of each pixels lies in candidate
region Each pixel is represent as Cregion
PRS factor is fixed and it was normally 0.5,0.4&0.3 After detecting region, the region will replaced by vertical
lines.
LP REGION
CODE FLOW CHART
MAKING A VOTE
Column with top and bottom neighbor have high blackness ratio will give a vote.
After voting section, the candidate region which have highest vote will be selected.
Finally plate will be detect and extracted.
EXPERIMENTAL SETUP
Web camera should be in live condition. 2-4 meter distance.
EXPERIMENTAL CONDITIONS
CLASSIFICATIONS
IMAGES
RESULT AND COMPARISON
Accuracy is higher than other LPD and algorithm useful for real time application
Computation time of each stages
Comparing with existing system
CONCLUSION
Using web camera is for monitoring vehicles and also low resolution images are used
New and fast algorithm which is useful for real time
requirements
Computation time is of 47.7 ms with an efficiency of 91.4%
Five to nine times faster than existing system
REFERENCES
License plate recognition (LPR) technology : impact evaluation and community assessment for law enforcement
A Real-Time Mobile Vehicle License Plate Detection and Recognition; Kuo-Ming Hung and Ching-Tang Hsieh
Comparison of feature extractors in LPR; S N Hinda,K Marsuki,Y Rubiyah,O Kharuddin
www.wikipedia.com