Traditional Photogrammetry
Image-based methods can be considered as the passive version of SL. In principle, image-based methods involve stereo calibration, feature extraction, feature correspondence analysis and depth computation based on corresponding points. It is a simple and low cost (in terms of equipment) approach, but it involves the challenging task of correctly identifying common points between images. Photogrammetry is the primary image-based method that is used to determine the 2D and 3D geometric properties of the objects that are visible in an image set.
The determination of the attitude, the position and the intrinsic geometric characteristics of the camera is known as the fundamental photogrammetric problem. It can be described as the determination of camera interior and exterior orientation parameters, as well as the determination of the 3D coordinates of points on the images. Photogrammetry can be divided into two categories. These are the aerial and the terrestrial photogrammetry.
In aerial photogrammetry, images are acquired via overhead shots from an aircraft or an UAV, whilst in terrestrial photogrammetry images are captured from locations near or on the surface of the earth. Additionally, when the object size and the distance between the camera and object are less than 100m then terrestrial photogrammetry is also defined as close range photogrammetry. The accuracy of photogrammetric measurements is largely a function of the camera’s optics quality and sensor resolution. Current commercial and open photogrammetric software solutions are able to quickly perform tasks such as camera calibration, epipolar geometry computations and textured map 3D mesh generation. Common digital images can be used and under suitable conditions high accuracy measurements can be obtained. The method can be considered objective and reliable. Using modern software solutions it can be relatively simple to apply and has a low cost. When combined with accurate measurements derived from a total station for example it can produce models of high accuracy for scales of 1:100 and even higher.
Overlapping area of images captured at A and B are resolved within the 3D model space to enable the precise and accurate measurement of the model.
Semi Automated Image Based Methods
In recent times, the increase in the computation power has allowed the introduction of semi automated image-based methods. Such an example is the combination of Structure-From-Motion (SFM) and Dense Multi-View 3D Reconstruction (DMVR) methods. They can be considered as the current extension of image-based methods. Over the last few years a number of software solutions implementing the SFM-DMVR algorithms from unordered image collections have been made available to the broad public. More specifically SFM is considered an extension of stereo vision, where instead of image pairs the method attempts to reconstruct depth from a number of unordered images that depict a static scene or an object from arbitrary viewpoints.
Apart from the feature extraction phase, the trajectories of corresponding features over the image collection are also computed. The method mainly uses the corresponding features, which are shared between different images that depict overlapping areas, to calculate the intrinsic and extrinsic parameters of the camera. These parameters are related to the focal length, the image format, the principal point, the lens distortion coefficients, the location of the projection centre and the image orientation in 3D space. Many systems involve the bundle adjustment method in order to improve the accuracy of calculating the camera trajectory within the image collection, minimise the projection error and prevent the error-built up of the camera position tracking.
Diagram illustrating the principles of structure from motion (SFM) measurement from multiple overlapping images.
