As title, I tried to SIFT, ASIFT algorithm, the matching is wrong, this kind of feature point matching method using feature point gradient around 128 dimensional vector as the description of the operator, the match, but two of my image shooting Angle difference is very big, one is the aviation image tilt photography Angle (sky), is a camera phone image (with) standing on the ground, the same SIFT feature points, the two images on the surrounding pixels, the 128 - dimensional vector is different also, the results matched points to is wrong (below picture),
Problem: the images of the shooting Angle vary widely how to match? Through the contour line can realize match? Look at some of the literature based on line matching, did not find suitable for shooting Angle vary widely, or have any other better way? He has struggled for months, for the great god give advice or comments!
Mobile video:

Tilt aviation photogrammetry image:

CodePudding user response:

Pure algorithm of machine learning, and such, I think, is not very good to
It is best to deep learning a large number of samples

CodePudding user response:

First picture view differ so big, people is not easy to judge

CodePudding user response:

reference 1st floor shiter response:

machine learning, and such, I think, the pure algorithm is not very good
It is best to depth study of a large sample

thank you for your reply! Alas is on machine learning is not understand, mentor has been using sift this kind of feature matching doubt whether the feasibility... Back to tutor advice about machine learning method

CodePudding user response:

refer to the second floor SSBQRM response:

first picture view differ so big, people is not easy to judge

thank you! The perspective of the poor is too big, so commonly used feature point matching algorithm is bad to use, can cut so wonder is there any other Angle, such as geometry information...

CodePudding user response:

Sift method generally no more than 15 degrees, the general need information modeling, multiple points of view can see slam algorithm

CodePudding user response:

Unless there is the perspective of information, or not feasible degree, or pay treasure red envelope is not so difficult to find the

CodePudding user response:

reference 6 LouXiaoLi 1991 response:

unless there is a different Angle of view of information, or not feasible degree, or pay treasure red envelope is not so difficult to find the

there are five Angle (monitored before and after) images, a dozen zhang stood on the facade, different position and Angle cell phone photos, the two can be calculated respectively each point cloud, but the aviation image point cloud is absolute coordinates, mobile point cloud is relative coordinates, don't know how to match up the two point cloud, the ipc who tried not line, so I want to, from the perspective of the image matching, the image also can not find the matching relation between

CodePudding user response:

refer to 7th floor huachen0917 response:

Quote: reference 6 LouXiaoLi 1991 response:

unless there is a different Angle of view of information, or not feasible degree, or pay treasure red envelope is not so difficult to find the

there are five Angle (monitored before and after) images, a dozen zhang stood on the facade, different position and Angle cell phone photos, the two can be calculated respectively each point cloud, but the aviation image point cloud is absolute coordinates, mobile point cloud is relative coordinates, don't know how to match up the two point cloud, the ipc who tried not line, so I want to, from the perspective of the image matching, the result image also can not find the matching relation between

Camera coordinate change can coordinate matching,,,

CodePudding user response:

Want to ask next host problem solved now? If can solve the roughly describe the train of thought, I now also is met the problem need to be matching Angle vary widely, with more than a month,

CodePudding user response:

Asift know

CodePudding user response:

I want to ask, have you solved the problem, I also had the same problem now, saw a lot of literature, they are not specific to write, mobile phone photo and photos can be fusion model of the unmanned aerial vehicle (uav)

CodePudding user response:

Hello, please you solved the problem that I met the same problem