Learn something new every day
More Info... by email
A computer vision test presents a series of challenges to a vision algorithm and notes responses. The development of computer vision supports activities like automated image processing, patient diagnosis, and robot movement. Facilities with an interest in this subject use testing to determine the level of performance they can achieve with various algorithms and programs. This can help them determine where their work needs improvement, and what kinds of improvements to enact to make an algorithm more functional.
Like the human brain, a computer can act as a processor for visual information, with the use of cameras for visual input. Computer vision can vary from relatively simple processes like recognizing a specific item in the visual field to more complex analysis. This is done through programming as well as training, which involves computer vision testing to challenge programs. A lab is usually needed for a computer vision test to control variables and access high speed processing equipment.
In a computer vision test, the algorithm can be presented with a series of challenge images. These may vary in complexity, and can include reference and test images as well as targets, to see how it responds to the mixture as a whole. For a facial recognition program, for example, programmers want the computer to spot human faces, and not to get confused by things that might look like faces, such as a photograph of an oddly-shaped rock. The testers program the computer to offer an output, like circling a face or illuminating an indicator light, in response to the visual input.
Still images are not the only thing that can be used in a computer vision test. Computers can also work with video and live real-time events. They may need to be able to track specific targets in motion and perform a variety of operations. For example, the sighting and targeting systems in military aircraft can follow a target and automatically update trajectories and other parameters for the benefit of the pilot. More peacefully, live image tracking can be useful for people like sports photographers, who may rely on rapid auto-focus features when working with speedy subjects like racehorses.
A variety of tests can be used to push a program to the limit. As the testers identify weak points, they can make adjustments to the program and re-test it. Algorithms capable of learning can be critical for these kinds of activities, as the program can get more intelligent with each computer vision test. It learns from its mistakes and files this information away for future reference, to minimize the chance of false positives or negatives.