2015-09-13 21:39:52 -0600 | asked a question | Homography of a Rectangle Hello. I am attempting to map a model to an image. The homography transform seems to be required, however i do not have corresponding features. Using colour segmentation, it is possible to get a binary image. We know which corner we are viewing which gives us one model point, and that the angle should be 90 degrees. We could extrapolate the other corner using hough lines.... I want to use findHomograpy(scene, object, 0) but i am unsure how to find enough points. Use default least mean squares. Can this be done with two points? Kind regards, Daniel |
2015-05-24 00:56:48 -0600 | commented question | Error: Expression: _BLOCK_TYPE_IS_VALID(pHead -> nBlockUse) Same error with Chamfer matching sample. Has anyone made any progress? |
2015-05-24 00:56:22 -0600 | received badge | ● Supporter (source) |
2015-05-23 00:38:18 -0600 | asked a question | Multiple Objects Features2D Hello, I am doing the basic object detection using features2d as outlined in the tutorial. How can this be extended to multiple objects? Since the detection of features will occur in the input image that may have duplicate features, then how do we find two distinct objects? The problem is that with a BruteForce matcher, there is no guarantee that the features will align correctly. Scenario. Image contains two objects. Feature detector finds keypoints Descriptors are extracted. We now want to match the descriptors to the template image. The descriptors contain two sets of object features. I feel like the centroid of the object is important here, and that something like the CMT tracker would be a good way to go. Any tips? |