Object recognition
Object recognition, as part of computer vision, is an important feature in both augmented reality and virtual reality. AR uses object recognition to deliver contextually aware information and multimedia. In VR, many systems such as Chaperone of SteamVR are designed in such a way that they are aware of their surroundings, which is essential for safe operation. As VR and AR technologies become more mainstream, we will see a dramatic improvement in object recognition. These improvements will have to overcome, among many other things, varying lighting conditions, changes in shape and size, and work fast enough to be usable in everyday situations.
How does object recognition work?[edit]
There are two main methods to conduct object recognition in AR, appearance-based methods and feature-based methods. Each method contains several different sub-methods.
Appearance-based methods[edit]
As their names may suggest, appearance-based methods consider comparable and detectable features of objects and their similarity to templates or exemplars. The main challenge for this method is the simple fact that any single object may look completely differently depending on lighting conditions, distance or angle from which it is viewed, and even its age. This means that highly effective appearance-based system needs to have a large base of templates to use as a base of its evaluation algorithms, which presents obvious problems with storage space, time, and manpower necessary for accomplishing this task. The most commonly used approaches for appearance-based evaluation are edge matching, divide-and-conquer search, greyscale matching, gradient matching, histograms of receptive field responses, and large model bases.
Feature-based methods[edit]
Feature-based methods look for similar features in an imagined or ideal object and a real image. When we consider, for example, a face recognition, it is possible to program a set of features that are associated with the human face. Using these features, a software algorithm can generate a model that will be placed over the captured image. If some features of this object match the image we have a positive match. Common feature-based detection methods include interpretation trees, hypothesize and test method, pose consistency, pose clustering, invariance, geometric hashing, scale-invariant feature transform method, and speeded up robust features (SURF).