Question

For my final year project I’m doing a vehicle detail modification system. the system should be able to these following tasks. I’m using a size image of a vehicle taken from a fixed distance lets say from 5m. and different colors and rim images are stored I’m my application, that is is the basic idea.

1. detect the tire and the alloy rim of the vehicle
2. detect and measure the rim size of the installed rim
3. apply a new alloy rim to the vehicle ( embed or put the alloy rim on top of the original image haven’t decided yet)
4. change color of the vehicle body
5. apply tint colors to the vehicle windows

Measuring wheel size

Modified alloy wheel changed Image

so what would be my research areas for this project? what should i be aware of? can you guys please help me out?

i know i need to be using Photogrammetry techniques for the measurement of the wheels.

Answer 1

I think tasks 1-3 are likely more than enough for a final year project! Having a camera at a fixed distance won’t let you determine (I don’t believe) the size of objects in the image unless you also know some information about the internal parameters of the camera (focal length, etc.) Without that information you won’t be able to translate a pixel length into a reliable real-world measurement. That being said, you could just run some tests and as long as you kept the distance from the camera to the rim exactly the same between tests and made sure the camera was square to the vehicle, you may be OK there.

I would say that detecting the rims would be a challenge because rim styles are all going to be different. You can take advantage of the fact that the colour will generally be non-existent (look for pixels with very little saturation). However, that will include pixels in the car body if the car is white, black, grey, etc. Then you’ll maybe need to take advantage of the edge content within the rims to give you a clue as to where their extents are. Finally, you could use the Generalized Hough Transform to attempt to find ‘circular’ objects. Combining those three sets of data may give you a decent clue as to where the rims are. Once you know the boundaries of the original rim, getting its physical size and applying the new alloy rim image is fairly straightforward assuming you have some knowledge of the camera parameters as stated above.

If you simply have to find the vehicle body, I’d simply look for large areas in the image that are the same colour. Try to build a colour model to pick out what colour the vehicle must be. Then you’ll just have some problems if there are parts of the vehicle in shadow or in strong light and knowing how to automatically connect those pixels into your car body model.

Once you have a colour model for the entire vehicle, you may be able to determine windows by finding self-contained blocks of colour (or non-colour, as the case may be) that are completely within the car body.

These are just some ideas. You’ve already got the keywords you need (image segmentation for car body determination, edge detection possibly for rim detection, photogrammetry and camera calibration to be able to take real-world measurements, etc.) I think you should definitely attack the problem in pieces, because each component you’ve mentioned would be fairly difficult in its own right!

Good luck.