Image-Registration

Image registration is a digital image processing technique that helps us to align different images of the same scene.

Generated SLAM Maps

Estimating the quality of a SLAM algorithm means estimating the accuracy of its results, which can be determined by the map of the mobile platform created while driving.

[[https://github.com/Dhaour9x/Image-Registration/blob/master/images/GroundTruth.jpg | Ground Truth Map]] [[https://github.com/Dhaour9x/Image-Registration/blob/master/images/gmappingSLAM.jpg | Gmapping SLAM Map]] [[https://github.com/Dhaour9x/Image-Registration/blob/master/images/hectorSLAM.jpg | Hector SLAM Map]]

Since the SLAM resulting image for the map does not always have the correct dimensions (approximately the same as the Ground Truth Map), the dimensions of the generated SLAM maps must be corrected approximately to the dimensions of the Ground Truth Map by image registration with OpenCV. We also have to convert from .pgm to .png or .jpg.

SLAM	Ground truth	Gmapping SLAM	Hector SLAM
Width(px)	527	384	2048
Height(px)	452	384	2048

Now we want to align a certain image to the same angle as a ground truth image. In the images above, the first image can be considered a "reference image", while Gmapping and Hector SLAM images are not well suited for comparison. The image registration algorithm helps us to align images of Gmapping and Hector SLAM in the same plane as the ground truth image.

The Results after Registration

|

The image registration algorithm works as follows:

• Both images are converted to grayscale.
• Adjust features from the image to be aligned to the reference image and save the coordinates of the corresponding keypoints. Keypoints are simply the selected few points used to calculate the transformation (generally points that stand out), and descriptors are histograms of image gradients to characterize the appearance of a keypoint. In this algorithm we use the ORB (Oriented FAST and Rotated BRIEF) implementation in the OpenCV library, which provides us with both the keypoints and the associated descriptors
• Select the best matches and remove the noisy matches.
• Find the homomorphic transformation.
• Apply this transformation to the original unaligned image to obtain the output image.

Installation:

git clone https://github.com/Dhaour9x/Image-Regidtration.git
cd Image-Registration

Usage:

python imRegister.py --first images/GroundTruth.jpg --second images/gmappingSLAM.png

---

@author Riadh Dhaoui