Is it possible to implement the identification numbers on the mobile device using a neural network?

Write a thesis project, you must implement a mobile application, such as recognition of car numbers, but will be recognized by the number of cars. The difficulty lies in the localization region on the photo containing the number. In the case of car numbers, location of precinct number is usually done by searching in the picture of rectangular outline which is missing on the cars.

I understand that these problem can be solved using neural networks. But then the question arises about performance. The app should work without a connection to the server. How quickly the neural network will work on mobile devices, whether to use it or is there some other solution to this problem?
September 19th 19 at 12:26
1 answer
September 19th 19 at 12:28
And have pictures? You can put a couple of examples?
Time of the network depends on the structure, but expect it to be O(d * n^2), where d-the depth, n is the size of the input.

UPD. Of course, I expect that the training of the network to occur on a more powerful device than the phone.

Find more questions by tags Neural networkspattern recognitionMobile development