Simple Marker Detection on Android (not traditional augmented reality), possibly with OpenCV

Question

For my final year project at university, I am extending an application called Rviz for Android. This is an application for Android tablets that uses ROS (robot operating system) to display information coming from robots. The main intent of the project is essentially the opposite of traditional augmented reality - instead of projecting something digital onto a view of the real world, I am projecting a view of the real world, coming from the tablet's camera, onto a digital view of the world (an abstract map). The intended purpose of the application is that the view of the real world should move on the map as the tablet moves around.

To move the view of the camera feed on screen, I am using the tablet's accelerometer and calculating distance travelled. This is inherently flawed, and as such, the movement is far from accurate (this itself doesn't matter that much - it's great material for my report). To improve the movement of the camera feed, I wish to use markers placed at predefined positions in the real world, with the intent that, if a marker is detected, the view jumps to the position of the marker. Unfortunately, while there are many SDKs out there that deal with marker detection (such as the Qualcomm SDK), they are all geared towards proper augmented reality (that is to say, overlaying something on top of a marker).

So far, the only two frameworks I have identified that could be somewhat useful are OpenCV (which looks very promising indeed, though I'm not very experienced with C++) and AndAR, which again seems very focused on traditional AR uses, but I might be able to modify. Would either of these frameworks be appropriate here? Is there any other way I could implement a solution?

If it helps at all, this is the source code of the application I am extending, and this is the source code for ROS on Android (the code which uses the camera is in the "android_gingerbread_mr1" folder. I can also provide a link to my extensions to Rviz, if that would also help. Thank you very much!

Edit: the main issue I'm having at the moment is trying to integrate the two separate classes which access the camera (JavaCameraView in OpenCV, and CameraPreviewView in ROS). They both need to be active at the same time, but they do different things. I'm sure I can combine them. As previously mentioned, I'll link to/upload the classes in question if needed.

Answer 1

So I've managed to find a solution to my problem, and it's completely different from what I thought it would be. All of the image processing is offloaded onto a computer, and is performed by a ROS node. This node uses a library called ArUco to detect markers. The markers are generated by a separate program provided with the library, and each has its own unique ID.

When a marker is detected, a ROS message is published containing the marker's ID. The app on the tablet receives the message, and moves the real-world view according to which marker it receives. It works pretty well, though it's a bit unreliable, because I have to use a low image quality to make rendering and transport of the image quicker. And that's my solution! I may post my source code here once the project is completely finished.

Answer 2

Have a look at the section about Template Matching in the OpenCV documentation. This thread may also be useful.