ProTrack's products are based upon proprietary technology developed in the company. The core technology consists of two major components:
Image Matching (Registration).
This technology makes it possible to match two or more misaligned images of the same scene. When presented with two images of a scene which are not aligned, i.e., one image having been rotated, shifted or zoomed relative to the other, the technology makes it possible to compute the parameters needed to match the second image to the first, subsequently yielding a precise, coherent joint image.
The technology uses an innovative probabilistic algorithm, much different from commonly used methods of matching, making it possible to gain a match, faster and more accurately then by other methods, even under difficult circumstances. Used with a standard PC computer, it may reach a matching rate of 60 video images per second with sub-pixel accuracy, of images differing from each other by zoom, shifts, rotations or even full affine transformation (6 parameters linear transformation).
A match may be gained even when the two images are not completely similar, such as when taken under different weather condition, produced with different types of cameras, or with different types of illumination. Unlike currently existing products, the algorithm works well even in the presence of considerable image noise,turbulence, as well as other types of interference. Furthermore, the top existing products rely upon very expensive special purpose hardware, without approaching the full ability of our technology.
One application of this capability is in video stabilization. A video camera held with trembling hands, or mounted on a moving vehicle, airplane, or helicopter |yields a rather unstable, blurred video picture, especially in zoom-in mode. Our technology may be used to stabilize the picture, by matching each image in the video sequence to the previous one, thereby canceling the tremors. It is possible to remove the effects of vibration, rotation, as well as of changes in the location of the camera in space.
Another application entails the matching of one or more images of small portions of a scene to their proper place in a big panoramic view of that scene.
A further application involves the matching of two aerial photos of the same scene, each taken with a different camera from a different position under different weather condition, a quite common situation.
Motion Detection.
The second core technology involves software for the detection of moving objects by means of a video camera or any other source of video picture. The technology makes it possible to detect moving objects appearing in the video picture. Many versions of simple motion detection software are available on the market, but none possess the capabilities and high performances of ProTrack's unique technology.
Our innovative algorithm is based on a set of virtual motion-sensitive detectors spread over the image. Each detector tracks a tiny portion of the image, and when such motion is detected, it is turned on. It is possible to define filters that limit detection to a specific direction, as well as ranges of speed and displacement. These filters lead to a powerful motion detection system that is highly sensitive to very small moving objects, on the one hand, but highly resistant to false alarms.
A filter may, for example, be tuned so as not to detect the swaying motions of a tree in the wind but still detect persistent motion in a distinct direction.
ProTrack's motion detection algorithm may be combined with its image matching technology, making it possible to detect motion even when the camera, rotates, vibrates, or moves, whether randomly or intentionally while scanning an area. This combination would, likewise, be insensitive to changes in illumination conditions caused by clouds, automatic adjustment of the camera focus and diaphragm, or any other reason.
The virtues of ProTrack's motion detection technology described above are a consequence of the unique approach to the problem, giving us a significant edge over existing algorithms.
|
