Machine learning Made Easy !!!/Image

Here we show the sample of Speed Recognition in Signboard using MATLAB.

In Basic Image Consider as Matrix, it’s have a values for Color Images like R, G, B.Single Value for gray-scale Images.

From figure we have easily understand the process. we have explain little bit with MATLAB code.

Overall Architecture

It is Have simple steps like 1. Pre-processing 2. Region of Interest/Circle Detection 3. Feature Extraction 4. Classification

Input Image

Step 1: Preprocess (src_img)

R2 =double (src_img (i, i, 1));

G2 =double (src_img (i, i, 2));

B2 =Double (src_img (i, i, 3));

Ohm RB=max(R/(R+G+B),B/(R+G+B))

Normally Image as 8 bit unsigned integer format. we need to convert double for mathematical operation.

f = im2bw (ohm RB);

f = bwareaopen(f,50);

simply apply binary segment with morphology open operation

Pre-Processed Image

Step2: ROI(Circle Verification)

IC = max(sqrt((xi-xc)2 +(yi-yc)2)- R)

if(IC<6.0) //approximatically

circle

else

No Circle Objects Detected

Since

a) (xi-xc)2 + (yi-yc)2 = R2

Region of Interest Verification

Step 3: Feature Extraction

It’s nothing but extract may be intensity, texture, shape, color related feature value.Here we use texture properties using Histogram and gradients.

HoG Features

Step 4: SVM Training

It’s a classifier for easy understanding:

Y=Wi*X + bi;

Wi =Weight Value

X =Feature

bi =bias value

Training is simply identify weight value of Wi & bi

We know Wi & bi from that we easily identify X.

Step 5: Classifier (SVM)

Support Vector Machine

Simple SVM

Step 6: Result

Classified Result as 60 Km/hr,20 Km/hr or No Detection

Option:

It’s easily connect with video as i/p and connect through embedded unit as output.

Link:

The Source Code Updated in Below link

amburosesekar/Machine-learning-Speed-Recognition

Thank you For Reading :)