Revision history [back]

I'm trying to generate a mosaic from a sequence of overlapping images.

I thought that a possible approach could be to loop over each image, find matching points and mosaic the first 2 images, then use the resulting mosaic as input array and find the matching point with the next image... and so on until the last frame. Pseudo code:

def mosaicing(imagesequence):
    matchingpoints = {}
    #
    # imagesequence can be a list of numpy array
    # 
    mosaic=imagesequence[0]
    for i in imagesequence[1:]:
        xy = find_matching_points(mosaic, i)
        # xy could be a list of x and y values for the matching point between 2 arrays
        # something like: 
        # matchingpoints[image_12] = [  [[x1,y1],[x2,y2], ... [xn,yn]] ,
        #                               [[x1,y1],[x2,y2], ... [xn,yn]]  ]
        # where the first array stores the pixel position for the mathcing points in image_1    ,       
        # and the second  array stores the pixel position for the same points  in image_2     
        #                                         
        mosaic = imagestiching(mosaic, i, xy) 
        # where mosaic is the first image and "i" is the second image
    return mosaic

In my use-case each image has an overlap of ~40%, they are collected along a straight line path with minimal pitch/roll, so I guess a rigid affine transformation based on 2 or more matching points should work fine. I stored 4 images sample (png) at [1]

Have you any clue on how to do this using python and opencv? I'm looking for a solution to:

1. detect matching point between 2 images (numpy nD arrays)

2. apply matching point to the 2 arrays to generate a new one which can be then used as input to detect the other matching points with the next frame

[1] http://epinux.com/test_images/