The Orange Objects Classification
My process Get the image from the robot Mark the position of the ball in each image Calculate all 7 attributes –Width –Height –Area –Perimeter –Form Factor –Distance –Eigen Value
Original Image
Mark the position of the ball
HUE color model
Orange objects in HUE
Discard Big objects
Find interested attribute Width, Height, and Area Width Height Area
Perimeter = ( * # of Diagonal lines) + # of Straight lines From this image: # of Diagonal line = 45 lines # of Straight line = 80 lines Perimeter = Find interested attributes
Perimeter = 18* +28 = WHAPFF.EigDist.APFF.EigDist.
Find interested attributes Form Factor (FF.) = where A = area, and P = perimeter If FF. is or close to 1 the circle object
Form Factor (FF.) BlueYellow Area121 Perim.4044 FF
How to calculate the perimeter
= 1 diagonal line and 1 straight line
= 1 diagonal line
= 2 diagonal lines
= 1 straight line
= 2 straight lines
Old Eigen Value We set the threshold value as EigVal 1 > (0.75*EigVal 2 ) It’s a ball
New Eigen Value (let C4.5 learn it) Eigen Value ratio (Eig) Eig = 1 st Eigen value / 2 nd Eigen value Eig
Find interested attributes Centroids and Distance
C4.5 result 1 st combination 2 nd combination New training DataNew test Data Training DataTest Data
Decision tree (7 attributes) size after pruning: 59 nodes <= > <=29>29 >14 <= <=0.4753> <=0.7965> <=64>64 <=0.5609>0.5609<=4468 >4468 <= > <=10 >10 <=0.8379> >13<=13 <=14 <=0.8812> > <= <=0.5262> <=0.6013> > <=16>16 <= > >0.9053<= > <= <=9>9 <= > <=13>13 >85<=85 <=80>80 <= > <= > <= >
Conclusion Gained more experiment with machine learning, C4.5 More practice with Matlab
Perimeter = 18* +28 = WHAPFF.EigDist.APFF.EigDist.