Can anyone recommend an approach to specify a subset of a frame for face detection? This could be for a fixed area, say one quadrant of the frame, throughout the face tracking - so not dynamic per frame. In my case I only need to scan a fixed part of the frame and am interesting in improving performance.
Limiting tracking to prominent-face-only is not an option as I need to track multiple faces and cropping each frame before sending to detector is too slow. My hope is to achieve something like this:
// Send entire frame, but specify subset for detection
detector.receiveFrame(fullFrame, sub_x, sub_y, sub_width, sub_height)
Or when creating a frame:
// Create a partial frame from a full frame during frame building
Frame.Builder().setImageData(fullFrame_byteBuffer, sub_x, sub_y, sub_width, sub_height, format)
Am I missing an option to do this within the SDK, or might there be a better approach than doing a copy/cropping operation on the full frame?
Thanks
Update: I understand that this feature is present in the underlying Pittpatt implementation as ppr_set_search_region(context, x1, y1, x2, y2) - Might Google make this available on Android?
This is not currently possible. While this answer is in response to dynamically limiting the detection area for bar codes, the answer seems to state categorically that limiting detection area is not currently supported.