Image Object recognition

[hmha]

hi,
the requires of this project are clasifing two type of vihicle: small and lage, but in the real conditions: lighting, backround change; other object moving;...
anyone give my the tips.
thank.

 

[leekk8]

If the background is consistent and only want to justify the size of the vehicle, you can try to substract the photo with vehicle with a photo without vehicle. After substraction, non-zero area should be the vehicle, and you can define the size by calculating the number of pixels of non-zero.

Just my thinking...

 

[hmha]

If the background is consistent and only want to justify the size of the vehicle, you can try to substract the photo with vehicle with a photo without vehicle. After substraction, non-zero area should be the vehicle, and you can define the size by calculating the number of pixels of non-zero.

Just my thinking...

thank leek,

the problems are: the shadow of the of objects vs. time is not same, the boundary of image object is not size of vehicle, and the neighbours object,...

 

[raylito]

You need the camera to always be fixed in one orientation. Then compare a picture with no trucks in, to one with trucks in by subtracting the one from the other. Use a loose comparison and you will find that most of the pixels that have changed, belong to the truck.

Use a number of different background pictures with different lighting shading etc..

Or you can use an edge detection filter. Basically subtract every [x+1,y] pixel from every [x,y] pixel for a horizontal edge filter. For a vertical edge filter subtract [x,y+1] from [x,y]. This will give you all the horizontal and vertical edges. Scan from the top down, and you will get the top of the truck. Bottom up and you will get the bottom of the truck.