All interested in, or have just started learning 3D reconstruction knowledge, will be more or less confused by the differences between the terms photogrammetry and computer vision. Some may suggest that the reconstruction of the point cloud is based on photogrammetry, while others insist on 3D computer vision.
It is important to know the differences between these subjects so that one can have a better understanding of these theories.
There are some shared characteristics between these subjects like they are both used to measure the 3D coordinate of an object, and both consist of well-developed theories and verification experiments. However, photogrammetry and computer vision differ from each other in 3 aspects: differences in the backbone theory, differences in the essence, and differences in the extended subjects.
1. differences in the backbone theory
The photogrammetry backbone theories are mainly constructed by the physical camera model and the survey theory, while in the computer vision community, they are more focused on the geometry model and linear transformation. For instance, in photogrammetry, the collinearity property describes the preservation of linear property among the object space and the image space, while this property is described in computer vision theory by the perspective transformation. Another example is that in the problem of geo-referencing, which is called point cloud registration in computer vision. The photogrammetry, which is highly based on survey theory, will focus on understanding the minimal observation and constraints needed to solve for all 7 unknown parameters (translation, rotation, and scale), while from the perspective of computer vision, it focuses on finding the correspondence features between the coordinate systems and apply linear algebra technique on solving the transformation. These deviations in backbone theory bring us to the second difference.
2. differences in the essence
Photogrammetry focuses on the quality of the solution, while computer vision focuses on efficiency. This can be seen from the algorithm they used. Photogrammetry often solves the equations using the least square fitting method, which includes an iterative calculation that is more accurate but time-consuming. However, computer vision algorithms often simplify the equation by linearizing them, thus obtaining a direct solution that is less accurate but more efficient. The differences in the focus sometimes create deviation between these two communities, but it is important to consider both aspects for real-world applications.
3. differences in the extended subjects
Photogrammetry discusses more about the procedure and the factors of obtaining a good 3D coordinate measurement, while computer vision extends more on the 3D scene understanding. In the photogrammetry community, more discussion focuses on the factors affecting 3D coordinate measurement, different camera models, camera calibration, or underwater 3D reconstruction. However, the computer vision community extends their research more to 3D scene understanding, like object pose estimation, point cloud object detection and segmentation, or integration with generative AI models.
In conclusion, there are 3 perspectives to understand the differences between photogrammetry and 3D computer vision, which are the backbone theory, the essence, and the extended subjects and applications. While these subjects are distinct from different perspectives, It is important to understand both of them because efficiency and quality complement each other, and both of them are important to real-world applications.
Thanks for reading! Can you think of more differences between the photogrammetry and the computer vision? if you like this post or have any suggestions or thoughts, please like and leave a comment!
Pixel Decoder 
Leave a comment