Reduce prediction from 6 seconds to 1 second

[josefermin]

The current prediction loop consist of three main steps:

1. Capture frame from video stream. (around 2-3 seconds)
2. Processing the image using dart (around 2-3 seconds)
3. Calling model.predict using the image as input (around 400ms)

Improvements:

• Remove step 2 completely. This step might be no necessary.
• Debug the code to understand why "Capturing frame from video stream" takes 2-3 seconds. An alternative solution could be to grab the screenshot directly from html.