FAST Corner Detector
This simple demo app takes the live camera feed, detects the FAST-9 corners based on the method by E. Rosten, and displays the results in a smaller window in near real-time.
FAST (Features from Accelerated Segment Test) can be computed very quickly. Great for real-time visual tracking. Of course, descriptors are not implemented in this project. If someone wants to do that, please feel free, and share your code with us! It seems that FAST9 corners have higher repeatability than the other ones (i.e. FAST12, FAST11, etc.).
In the v0.2 release, I added the non-maximal suppression stage to eliminate adjacent corners. These are shown as green dots, while the original detected corners are red dots.
You can now download the source code of Ztitch. The project has been simplified compared to the actual app. For example, there's no Facebook/Flickr services, or Bing maps access.
However, the basics which teach you how to render images in a spherical panorama, as well as the handlers for controlling navigation via swipe or accelerometer, are still contained within.
Note that the project includes four sample images as local content (pictured above), and their orientation have been hard-coded in the SetupImages() method. The UpdateImageMatrices() method then takes the supplied information to calculate a final perspective projection matrix for each image. If you use your own images, then you should update SetupImages() with your own code. There are many possibilities with this project. For example, if you can get a bundle adjustment algorithm working on the phone or as an online service, then you could automatically arrange images in a semi-3D environment not limited to a spherical panorama.