Augmented reality implementation method

A method for implementing augmented reality by transferring points of interest (PoI) from at least one model image to at least one destination image, the method comprising the step of creating a detection model starting from at least one model image on which at least one point of interest (PoI) is represented, wherein said step of creating a detection model comprises the sub-steps of:

• extracting a predefined group of keypoints (P1, Pn) from each model image;
• calculating a descriptor for each of the keypoints (P1, Pn) of said predefined group of keypoints (P1, Pn) extracted from said model image;
• for each point of interest (PoI), selecting a predefined subgroup of keypoints (P1, Pk) adjacent in spatial terms and selecting all the pairs (P1, P2) of said predefined subgroup of keypoints (P1, Pk) such that the following spatial relationships are valid:
• the distance (P1-P2) between a first keypoint (P1) and a second keypoint (P2) exceeds a first minimum threshold, and
• the ratio between the distance (P1-PoI) between a first keypoint (P1) and the point of interest (PoI) and the distance (P1-P2) between a first keypoint (P1) and a second keypoint (P2) is less than a second minimum threshold;
• calculating the angle value (P2, P1, PoI), measured at the vertex formed by the first keypoint (P1), between the segment joining the second keypoint (P2) and the first keypoint (P1) and the segment that joins the point of interest (PoI) and the first keypoint (P1);
• the method comprising the further steps of:
• identifying a specific model image of said detection model in said at least one destination image on the basis of the value of the distance, measured on the basis of said keypoint descriptors, between each keypoint (P1, Pn) of said at least one destination image and the nearest keypoint (P1, Pn) of said model image;
• once said specific model image has been identified, transferring said at least one point of interest (PoI) of said model image on said at least one destination image such that for each pair of keypoints (P1′, P2′) of said at least one destination image, which corresponds to a pair of similar keypoints (P1, P2) of said model image, a continuous probability distribution is made to identify the position of said at least one point of interest (PoI) on said at least one destination image through the values of said angle (P2, P1, PoI) and of said ratio between the distance (P1-PoI) between the first keypoint (P1) and the point of interest (PoI) and the distance (P1-P2) between the first keypoint (P1) and the second keypoint (P2), wherein the argmax of the combination of said probability distributions identifies the position of said at least one point of interest (PoI) on said at least one destination image;
• once the position of said at least one point of interest (PoI) has been identified on said at least one destination image, starting a tracker of points of interest (PoI) that tracks the position of said point of interests on subsequent destination images.