A few technical test that have been done were understanding the appropriate amount of track points to extract when converting an image into a natural feature set.
AR tool kit allows you to specify a few different parameters when it comes to customising the output result. The first parameter is DPI. Its important to note that the higher the DPI the larger the feature set and slower it is to load. It should also be noted that mobile camera’s have a maximum resolution (DPI) so it is redundant to go over this. With all of these factors and after testing the best DPI was 150.
The second parameter that can be defined is initialisation threshold. This has a range of 0 – 4. 0 meaning that only a few points have to be detected for NFT to load and track and 4 meaning that a high amount of points must be detected. I have found that a value of 1 works best for this. It allows the users a little bit of give in getting the scene initiated.
The third parameter is amount of track points. This also ranges from a value of 0 – 4. The best option for this varies on the image. As a general rule of thumb if an image has a lot of “noise” then a lower number of track points is recommended. However if an image is clean or has been digitally created then a setting of 3 is recommended. For the best results of “realworld NFT” I have found that digital clean ups then extracting at level 3 is the best. – see NFT clean up blog post for more info