Intersection Tie Points (ITP) are designed to be used in man-made environments including both interior and urban scenes. They are a type of machine learning object that uses intersecting linear features to generate a tie point. Similar to other tie points, ITPs are used to help calibrate the scene. However, they define intersection points that humans would consider, rather than clusters of unique pixels and image texture.
IN THIS ARTICLE
All tie points are integral for the calibration and reconstruction of the scene. Using ITPs can add additional influence during this process of calibration. However, not every tie point is considered equal during the calibration step, and understanding this relationship is important for an accurate reconstruction.
To begin, Automatic Tie Points (ATPs) are the result of an automated keypoint extraction process of unique pixels and image texture. As a result, ATPs are considered the least trustworthy. While still important, they are given the least amount of influence on the calibration of the scene.
ITPs are based on matching meaningful and recognizable linear entities within overlapping images. They come in two varieties: Automatic ITPs (aITPs) and Manual ITPs (mITPs).
aITPs are generated from a machine learning process of matching meaningful and recognizable linear entities within overlapping images. This machine learning process is considered more trustworthy than the automated keypoint extraction process from ATPs. As a result, aITPs have more influence on the calibration than ATPs.
mITPs are manually identified intersection points. Manually created and identified points, including mITPs and Manual Tie Points (MTPs), are considered the most accurate and thereby the most trustworthy. mITPs and MTPs have equal weight and have the greatest influence on the calibration of the scene.
|Tie point hierarchy
|ATP < aITP < (mITPs=MTP)
Automatic ITPs are automatically identified linear features that are matched in overlapping images during the process of image alignment. PIX4Dmatic identifies aITPs by extracting linear features such as a corner of a room, a window, or perhaps a door. These features are symbolized as green line segments in the Image viewer. These line segments are extracted per image, and later matched between images in order to create an intersecting tie point.
aITP in Image viewer.
Simply enable the Automatic ITP toggle in the Calibration step.
Automatic ITP toggle.
Manual ITPs are similar to traditional MTPs. mITPs and MTPs are unique points that are manually identified and marked by the user in at least two overlapping images. With MTPs the user must identify unique clusters of pixels or textures within overlapping images. However, with mITPs the user must manually identify at least two linear features within an image that intersect. These linear features are symbolized as yellow line segments in the images, and they help the user to better mark images by providing visual confirmation. If the marking of an mITP in an image is more accurately established, then calibration will have a better result. There are two methods for creating an mITP either by converting an aITP to an mITP, or creating an entirely new mITP.
mITP in Image Viewer.
An automatic ITP can be converted to a manual ITP. After the calibration step has been completed and aITPs have been generated, open the Image viewer and click on an aITP in the 3D view. The green automatic line segments will be displayed in the Image viewer. Clicking on the green line segment in an image will change the color to yellow, and thereby change the aITP to an mITP. Converting to mITP will give the point more influence during reoptimization.
aITP after Calibrate step
Same aITP converted to mITP
A new mITP can be created by opening the image viewer. Click on the Mark new mITP icon in the Image viewer . Begin by finding common line intersections between images and click at the intersection point. Move the line segments over the linear features to help establish the mITP. A minimum of two images are required to be marked. The image below displays a corner of a door in two separate images.
mITP marked in image one
mITP marked in image two
Both aITPs and mITPs have the same symbol in the 3D viewer.
ITPs in 3D viewer.