JNI, Accessing SurfaceTexture using Native Code

Question

I need to be able to execute native code ( algorithms ) on a video stream generated by the device camera, for that, I have considered OpenCV camera capture, Unfortunetly, at this time, it doesn't support Android 4.0.3 @ Samsung G2 which is my target device, as an alternative, I consider capture using the Java Camera object, and, using JNI, to ~Marshal~ the captured data to the native domain, this however, impose marshaling overhead ( coping the data ), to avoid that, I have considered rendering the captured image ( preview ) to a GL Texture ( using eg. Camera.setPreviewTexture ), and, directly accessing the GL Texture @ the native domain, and thus, avoiding the un-needed coping.

Is there a way to directly access the TextureSurface @ the native domain ?

Any help will be appreciated.

Nadav At Sophin

Answer 1

The VideoPlayback sample app released as part of Qualcomm's Vuforia augmented reality SDK achieves this I think. I've only just started going over the code myself these past few days and a lot of it's new to me so I'm not 100% sure.

https://ar.qualcomm.at/content/video-playback-sample-app-posted

If I'm right, the app plays a movie file through the Java domain's MediaPlayer class which is rendered to a SurfaceTexture which is then in turn accessed by the OpenGL ES code in the native domain to be rendered on the actual augmented reality display.