views:

3213

answers:

4

I am looking into an image processing problem for semi-real time detection of certain scenarios. My goal is to have the live video arrive as Motion JPEG frames in my Java code somehow.

I am familiar with the Java Media Framework and, sadly, I think we can consider that an effectively dead API. I am also familiar with Axis boxes and, while I really like their solution, I would appreciate any critical feedback on my specific points of interest.

This is how I define "best" for the purpose of this discussion:

  • Latency - if I'm controlling the camera using this video stream, I would like to keep my round-trip latency at less than 100 milliseconds if possible. That's measured as the time between my control input to the time when I see the visible change. EDIT some time later: another thing to keep in mind is that camera control is likely to be a combination of manual and automatic (event triggers). We need to see those pictures right away, even if the high quality feed is archived separately.
  • Cost - free / open source is better than not free.
  • Adjustable codec parameters - I need to be able to tune the codec for certain situations. Sometimes a high-speed low-resolution stream is actually easier to process.
  • "Integration" with Java - how much trouble is it to hook this solution to my code? Am I sending packets over a socket? Hitting URLs? Installing Direct3D / JNI combinations?
  • Windows / Linux / both? - I would prefer an operating system agnostic solution because I have to deliver to several flavors of OS but there may be a solution that is optimal for one but not the other.

NOTE: I am aware of other image / video capture codecs and that is not the focus of this question. I am specifically not interested in streaming APIs (e.g., MPEG4) due to the loss of frame accuracy. However, if there is a solution to my question that delivers another frame-accurate data stream, please chime in.

Follow-up to this question: at this point, I am strongly inclined to buy appliances such as the Axis video encoders rather than trying to capture the video in software or on the PC directly. However, if someone has alternatives, I'd love to hear them.

A: 

Have you ever looked at Processing.org? It's basically a simplified application framework for developing "artsy" applications and physical computing platforms, but it's based on Java and you can dig down to the "real" Java underneath.

The reason it came to mind is that there are several video libraries for Processing which are basically Java components (at least I think they are - the site has all the technical information you might need). There is a tutorial on using the Processing libraries and tools in the Eclipse IDE. There are also numerous examples on video capture and processing.

Even if you can't use the libraries directly, Processing is a great language/environment for working out algorithms. There are several great examples of image and video capture and real-time processing there.

CMPalmer
+1  A: 

Regarding the dead-ness of JMF, are you aware of the FMJ implementation? I don't know whether it qualifies as the "best" solution, but it's probably worth adding to the discussion.

Greg Mattes
The last update appears to have been quite a long time ago: http://sourceforge.net/project/showfiles.php?group_id=161490 An API that has been dead since 2007 is probably not going to help us. Thanks anyway, though.
Bob Cross
+1  A: 

i have no experience here but i wonder if this project may be helpful.. seems to have a focus on streaming so likely not

A: 

FMJ can definitely capture video and turn it into MJPEG frames.