Computer vision above the cloud

Question

Is there a way to make computer vision above the cloud? The idea is that people register on the website, then the webcam is activated, video data is sent to the server via the Internet. The server processes this data and sends the processed data to the user in real time or at least 10 frames per second.

Is this doable? What skills do we need on the network side? I know that streaming video is one of the components. Also, how can we configure the server? Can a distributed system help or not consider very large computations in a limited time?

Answer 1

Different levels of detection of large-scale space can be performed in parallel, and the database with which you are comparing your images can be distributed across several servers.

As I understand it, you want to create something like augmented reality. I can’t answer with an explicit yes or no, if it can be done with the current performance and bandwidth of the mobile processor.

I would start by implementing a very rudimentary detection of the function on the client side, and then sending the still images to the server (high resolution is the key). The server can process the image with high processing power and scan objects based on the database. Then submit the result.

The client can then hook up its most basic function discovery with the server’s response, and thus create a live video image. The server should be called when the client detects that new image data is available (the user turns the phone in the other direction).

Answer 2

It will be worth it if

1) you can compress your image data or functions sufficient to be viable with any bandwidth that the user has

2) the calculations that you do are large / complex, that they are not performed in the browser

If you determine that both of them are true, the easiest way to look is to send your functions or images via web servers to a server that is ready to classify them or perform all the processing you need. Perhaps look at the tornado websocket framework for python, then you can integrate with OpenCV bindings in python without any problems. Based on the information you gave, it’s hard for me to say much more.

Regardless of whether the distributed system helps, it depends on what you intend to do (which CV algorithm), but most likely it will be if you can implement it.

I would advise you to take a look at javascript solutions in the browser, because network latency will be a big problem.

Answer 3

See http://vision.ai/ . They launch kickstarter, which is an application for viewing applications with a thin client, where computer vision occurs on a remote server. They have object detectors, trackers, and other widgets and methods for teaching these features. Add them if you want this to happen.