What are some ideas for research projects?

What are some good ideas for computer science undergraduate research projects in the field of augmented reality?

• Answer:

  AR is still very much a technology that needs a lot of maturing. It is very impressive to watch movies like "Avatar" or "Iron Man" where AR plays a key role in the screenplay. It is an entirely different matter to try to realize this. A lot of complex technologies need to be developed before Science Fiction becomes "Science Fact". Tracking Virtual Reality (VR) is very easy to deal with because everything that you see is digitized. All objects that you see and manipulate are digital objects. This also includes their individual position in relation to each other. For AR this is immensely more complex since you need to relate digital information to real world imagery. Basically the problem is: where are you, and what are you looking at? On top of this you need to deal with latency of computer power. Try holding your Smartphone in your hand with the camera turned on. Now pan quickly around and notice that the image on the screen is "late". This is latency and is caused by the fact that it takes time for the phone to capture the image, process it in the CPU and display it on the screen. Now add a simple computer generated 3D box and attach it to an object in the real world and synchronize the motion of the 3D box to the real imagery on the screen. This sounds simple but is in fact one of the hardest thing to figure out in computer science. This is where tracking of the camera position is important. Add to this that it has to work together with latency and acceleration. That alone should give you headaches for an entire semester :-) Motion anticipation/prediction Let’s say that you now have tracking figured out. Now you need to figure out how to predict motions of camera AND 3D objects. Imagine that you hold a fiducial up in front of the camera and an object appears on the fiducial. When you move the fiducial in front of the camera you want to have the object following the motion of the fiducial. This way you can prepare calculation on how the object will look like in X latency time. This is especially important when you start working with transparent displays like "Google Glass". Here, reality has no latency which makes it crucial to have a solid motion anticipation system to encounter latency in the realized experience. It also brings up the question of motion anticipation for the camera position. When using transparent display it is extremely important to have a solid motion anticipation system for both the 3D objects and the camera position. Object Recognition. When you act in a mixed environment like AR it is not fun unless the computer has a way of recognizing what it is seeing. If the computer is unable to recognize what it sees it will not generate anything relevant to what you are looking at. Let’s say that you are looking at a piece of art in a museum. You would want the AR system to tell you what it is, who the artist etc. is. This calls for a solid Object Recognition system. It is also the object recognition system that places the data (3D model, text info) correctly in relation to the object. Together with tracking and motion anticipation you can start to actually use AR for something. Since you don't want to have a complete library of everything and everyone in world you need to customize the database to a specific need. For example: Exhibition in a specific museum, Limited technical installations or finding your way around the city. The keyword for all of this is "Speed". You need blazing fast tracking, motion anticipation and object recognition. If you fail on this, people will not use it in real life (except perhaps in a lab). ....Best of luck :-)