| Video image registration |
| Image Registration and Fusion Systems |
 | |  |
Introduction
Analysis of a video captured by a moving camera can be made simpler by registering the video frames at the background. Image
registration separates motion of the camera from motion of the objects, enabling detection and tracking of moving objects.
Image registration steps
A software developed by Image Registration and Fusion Systems registers consecutive frames in a video according to the
following steps:
Strength
The software is capable of separating the ground from elevated objects, such a buildings and trees, in an image and register the
images at ground level. This unique capability enable easy detection of objects that move on the ground. The software has a
number of parameters that can be adapted to different video types. The number of landmarks selected in each frame can be
adjusted. The number of correspondences used in image registration can be selected by setting the tolerance used to consider
points in the images as being physically the same. The software can register video frames as large as 10K by 10K pixels, and the
frames can be color or gray scale. The software can create a video of registered images in compressed form, or save the
transformation for each frame so that a video can be constructed later. For very large videos, the software can prepare the video on
disk (rather than in RAM) and display the video directly from disk. Registration is fully automatic. The user simply selects the first
and last frames in a video and the software will then register the frames to the first frame.
Input/Output
The software is currently set up to read jpg, ppm, pgm, and no header files. Other file formats can be incorporated into the system
upon request. A compressed video format developed at Image Registration and Fusion Systems is used to create videos that may
have very large frames.
Interaction
User interaction with the program is with the keyboard and mouse and options are available through a pull-down menu. The menu
enables selection of various operations of the software. The mouse enables interaction with the displayed image (e.g., scrolling
the image within the display window). When viewing a constructed video, the frames are advanced in forward or reverse direction
by pressing keys on the keyboard.
Speed
Speed of operation is dependent on image size and image content, both of which can be controlled by the number of landmarks
selected in each image. Depending on the image size, the user selects the desired number of landmarks. Number of landmarks
determines registration accuracy and reliability. To increase accuracy and/or reliability, the number of landmarks should be
increased. Typically, for a 1K by 1K frame, 100 landmarks per frame are sufficient, registering two frames in less than a minute.
Larger frames require proportionately more time.
Accuracy
Registration accuracy of the software can be controlled by adjusting two parameters: 1) the number of landmarks selected in each
frame and 2) the tolerance used to find correspondence between landmarks in consecutive frames. Obtained registration
accuracy also depends on frame rate of the camera as well as the distance of the camera to the scene and the variation in scene
relief.
Reliability
Reliability is considered one minus the rate of registration failure and it usually varies with registration accuracy. The higher the
accuracy, the higher the reliability will be. To achieve a highly reliable registrations, there is a need to set the parameters of the
system to obtain a highly accurate registration, which translates to a higher computational cost or a lower speed.
Enquiries
For questions or further information about this software, please contact cs@imgfsr.com.
To obtain a license for this software, follow this link =>
Analysis of a video captured by a moving camera can be made simpler by registering the video frames at the background. Image
registration separates motion of the camera from motion of the objects, enabling detection and tracking of moving objects.
Image registration steps
A software developed by Image Registration and Fusion Systems registers consecutive frames in a video according to the
following steps:
- Landmark detection: Selecting a set of landmarks from each frame. The user can specify the number of desired landmarks.
- Landmark correspondences: Determining correspondence between landmarks in the frames using pose consistency and
projective constraint. - Estimating transformation parameters: The parameters of a projective transformation are determined to match most
landmarks in the images. This transformation is made global so that all transformations are with respect to the first frame. - Resampling: Resampling the current frame to register with the first frame.
Strength
The software is capable of separating the ground from elevated objects, such a buildings and trees, in an image and register the
images at ground level. This unique capability enable easy detection of objects that move on the ground. The software has a
number of parameters that can be adapted to different video types. The number of landmarks selected in each frame can be
adjusted. The number of correspondences used in image registration can be selected by setting the tolerance used to consider
points in the images as being physically the same. The software can register video frames as large as 10K by 10K pixels, and the
frames can be color or gray scale. The software can create a video of registered images in compressed form, or save the
transformation for each frame so that a video can be constructed later. For very large videos, the software can prepare the video on
disk (rather than in RAM) and display the video directly from disk. Registration is fully automatic. The user simply selects the first
and last frames in a video and the software will then register the frames to the first frame.
Input/Output
The software is currently set up to read jpg, ppm, pgm, and no header files. Other file formats can be incorporated into the system
upon request. A compressed video format developed at Image Registration and Fusion Systems is used to create videos that may
have very large frames.
Interaction
User interaction with the program is with the keyboard and mouse and options are available through a pull-down menu. The menu
enables selection of various operations of the software. The mouse enables interaction with the displayed image (e.g., scrolling
the image within the display window). When viewing a constructed video, the frames are advanced in forward or reverse direction
by pressing keys on the keyboard.
Speed
Speed of operation is dependent on image size and image content, both of which can be controlled by the number of landmarks
selected in each image. Depending on the image size, the user selects the desired number of landmarks. Number of landmarks
determines registration accuracy and reliability. To increase accuracy and/or reliability, the number of landmarks should be
increased. Typically, for a 1K by 1K frame, 100 landmarks per frame are sufficient, registering two frames in less than a minute.
Larger frames require proportionately more time.
Accuracy
Registration accuracy of the software can be controlled by adjusting two parameters: 1) the number of landmarks selected in each
frame and 2) the tolerance used to find correspondence between landmarks in consecutive frames. Obtained registration
accuracy also depends on frame rate of the camera as well as the distance of the camera to the scene and the variation in scene
relief.
Reliability
Reliability is considered one minus the rate of registration failure and it usually varies with registration accuracy. The higher the
accuracy, the higher the reliability will be. To achieve a highly reliable registrations, there is a need to set the parameters of the
system to obtain a highly accurate registration, which translates to a higher computational cost or a lower speed.
Enquiries
For questions or further information about this software, please contact cs@imgfsr.com.
To obtain a license for this software, follow this link =>