Computer vision technology is a branch of artificial intelligence that focuses on providing computers with the functions typical of human vision.
According to researchers at the University of Sydney, Australia, human-centered applications for human-computer interfaces (HCIs), augmented perception, automatic media interpretation, and video surveillance are emerging. Here is one example.
The vOICe, developed at Philips Research Laboratories, provides a simple yet effective means of augmented perception for people with partially impaired vision. In the virtual demonstration, the camera accompanies you in your wanderings. The camera periodically scans the scene in front of you and turns images into sounds, using different pitches and lengths to encode objects’ position and size.
Now let's look at the Face Detection Project from Carnegie Mellon University (CMU).
The use of computer vision for automatic media interpretation assists users in searching for specific scenes and shots otherwise not annotated in the video-scene indexes. For example, images containing faces can be automatically distinguished from other images, as the results of the Face Detection Projectat CMU prove. The CMU face detector is considered the most accurate for frontal face detection and is also reliable for facial profiles and three-quarter images.
And of course, there is video surveillance. These researchers designed a system to track suspicious pedestrian behavior in parking lots. Here is a summary of their approach.
The first step consists of detecting all the moving objects in the scene by subtracting an estimated "background image" -- one that represents only the static objects in the scene -- from the current frame. The next step is to distinguish people from moving vehicles on the basis of a form factor, such as the height: width ratio, and to locate their heads as the top region in their silhouette. In this way, the head’s speed at each frame is automatically determined. Then, a series of speed samples are repeatedly measured for each person in the scene. Each series covers an interval of about 10 s, which is enough to detect suspicious behavior patterns.
Finally, a neural network classifier, trained to recognize the suspicious behaviors, provides the behavior classification. In the experiments we performed, the system achieved good accuracy, with a reasonably limited number of false dismissals and false alarms -- 4% and 2%, respectively, among more than 100 test samples.
For more information, figures and references, please read this excellent article.
Source: Massimo Picardi and Tony Jan, in The Industrial Physicist, February/March 2003, Volume 9, Number 1
6:25:23 PM Permalink
|
|
