Object Recognition is an aspect of computer vision that deals with the ability to find and identify individual objects from a scene or video clip containing one or more objects. This  is based on the premise that the computer is simply given a 'description' of the object and that it is able to apply this description to identify and localize that object from a group of objects.

The real challenge in object recognition is the ability to accomplish the task even if the object of interest is rotated, translated, scaled down or up, or even partially obscured.  Humans find the task of object recognition straight-forward, so it is the aim of computer engineers to understand why and develop a system that achieves the same level of object recognition in computer and robotic systems.

The description of an object to a computer may be model-based or appearance-based, or a combination of both.  Model-based descriptions define the geometric features of the object to represent it to the computer. On the other hand, appearance-based representation employs presenting a large set of images of the object to the computer in order to 'train' the computer to recognize it.

Object recognition itself comes in many flavors: recognition of a 2-D object from a 2-D image, recognition of a 3-D object from a 2-D image, recognition of a 3-D object from a 3-D image; recognition of a 2-D or 3-D image from a set of 2-D images taken from different angles; recognition of a 2-D or 3-D image from a video clip, etc.

A typical model-based object recognition system breaks down the task of object recognition into a series of steps such as: 1) acquisition of sensory data; 2) analysis of the acquired sensory data; 3) extraction of the features of the object; 4) organization of the extracted features; 5) construction of the model based on the organized features; and 6) matching of the perceived model with known models to identify the object.

See Also:   Computer Vision;  More Industry Articles