Hi Osman,

The distance threshold in ORB and other binary descriptors is used for determining if two descriptors are similar enough. When the hamming distance is computed, you obtain a value which determines the similarity between both descriptors. If the obtained value is less than the threshold, then the descriptors are matched.

Please, check the attached papers if you want to learn more about binary descriptors and its application in problems such as place recognition or loop closure detection.

Good luck with your research!

Conference Paper Bidirectional Loop Closure Detection on Panoramas for Visual...

Conference Paper Fast and Effective Visual Place Recognition using Binary Cod...

Thank you every one, I have found the answer, if you interested  check http://stackoverflow.com/questions/22857398/matching-orb-features-with-a-threshold

Osman my friend

there is no optimal answer (solution), in the link you provided.

the basic version of hamming distance works on the number of mismatches, for each two values if they mismatch it count s one, and so on for the rest of the array, normally, there is no threshold for the matching, although you may do one based on experiments.

however, Hamming distance is very sensitive to shifting, for example:

A={1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0}

B={0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1}

Ham(A,B)=16

while the Edit distance for instance= 2

I advice you to determine (through experiments) a threshold based on the overall distance not based on each point within the feature vector, I invented a similarity measure, which is invariant to data scale and outliers, so I am sure you will get better results. if you incorporate my distance with some classifier.

link to invariant similarity measure: https://www.researchgate.net/publication/264995324_Dimensionality_Invariant_Similarity_Measure

link to classifier: https://www.researchgate.net/publication/265168466_Solving_the_Problem_of_the_K_Parameter_in_the_KNN_Classifier_Using_an_Ensemble_Learning_Approach

if you want C++ code for both, I will send you upon your request.

Article Dimensionality Invariant Similarity Measure

Article Solving the Problem of the K Parameter in the KNN Classifier...

Thank your Mr hassanat, your opinion is convisible and you are very kind, I will try to read your paper.