I'd suggest you may want to consider doing the following:
- use the Cloud Recognition feature (you could start from the Cloud Reco sample for inspiration)
- encode a specific Video URL in each Metadata instance for each target; the video URL would point to some network location where your video file is actually stored (e.g. on your own server, for instance)
- in the application code, when a target serach result is returned, you could read the Metadata as a text string, and extract the "video URL" out of it
- use the video URL to load/stream the video as shown in the VideoPlayback sample code
Note: if you have some text too, you could embed the text again in the Metadata, i.e. the Metadata would contain 2 pieces of info: your text and the video URL.
Hope this helps get you started.