Experimental Orb feature

[kalwalt]

I'm trying to port the orb class form jsfeat_orb into webarkit-jsfeat-cpp. I will implement other classes that is required internally, but for a working example it will be necessary also other functionalities and classes not present in in the code at the moment.

list of feature implemented

• [x] keypoint_t class.
• [x] pyramid_t class.
• [x] imgproc warp_affine method.
• [x] imgproc resample.
• [x] imgproc pyrdown with example.
• [ ] ~imgproc gaussian_blur.~ This will be done in another step
• [ ] ~cache class (?) (used in many jsfeat classes).~ This will be done in another step
• [ ] ~matmath class (only some of the methods).~ This will be done in another step
• [ ] ~_math class.~ This will be done in another step
• [x] orb class ~with working example.~ orb class is implemented but not with a working example.
• [x] grayscale video example.

[kalwalt]

With the latest commit i have no errors in the console, but testing the sample_pyrdown.html example, fps are very slow. Maybe something wrong with my implementation?

[kalwalt]

I'm not sure that is required the cache class. For example if we look at the resample function https://github.com/inspirit/jsfeat/blob/4c7b336bbeeb26e6cd4cdf3c7d414abe273846f3/src/jsfeat_imgproc.js#L405-L417 it's derived from ccv library internally method resample_u8 is derived from ccv https://github.com/liuliu/ccv/blob/e058fb5d80a71c66bd39a49c4b06955c7631451c/lib/ccv_resample.c#L11 That is C code but we can do in C++ as well. Instead of allocating data with alloca we can use STL vector's in a feasible way and avoid creating a cache class.

[kalwalt]

I'm testing the resample function, as jsfeat is heavily inspired by the ccv lib i'm trying to convert the C code to C++. I haven't created bindings for emscripten but they will be soon available.

[kalwalt]

I think i will implement a lru cache in another Pr. I will try to follow this article https://www.nextptr.com/tutorial/ta1576645374/stdlist-splice-for-implementing-lru-cache.

[kalwalt]

The sample_orb_mixed.html not works completely, it can only train the pattern but will fail in the match_pattern routine, infact here screen_descriptors expect the old jsfeat syntax where you can var query_u32 = screen_descriptors.buffer.i32; // cast to integer buffer but with jsfeatCpp you can't. So i think the only way is to convert to the old syntax internally match_pattern, but at the moment we have not such copyTo or moveTo function inside matrix_t jsfeatCpp, so this will be done in another moment. I will merge this PR, leaving the tasks undone for another PRs. Now proority is to imporove matrix_t class:

• inheriting matrix_t from data_t class
• create some copyTo or moveTo functions to copy from jsfeat and not only.