This model takes as input two or more photos, along with available geospatial metadata, in JPEG, PNG, TIFF, or NITF encoded RGB format, no larger than 5000 x 5000 pixels. It has two outputs: 1) the geometric transformation homography that transforms the ‘aligned’ image to the ‘reference’ image, and 2) a new GeoTIFF registered image based on the transformation information. This model can be used in remote sensing to detect changes in many phenomena such as vegetation, habitation, buildings, and vehicles. It can also be used to track traffic patterns over time, as well as activity such as construction.
Create a Modzy account to get started →
Since this model is a preprocessing model for imagery, it does not have any associated and relevant performance metrics. The model tends to perform better on images with scenes of flat grounds (e.g., fields, tops of buildings) instead of scenes with significant contour (e.g., hills, areas with many trees, riverbanks).
This model was not trained on any dataset, as it is a preprocessing tool that uses a transformation methodology built into the OpenCV library. The algorithm first finds key-points and descriptors from each image using the ORB OpenCV function. It then matches the key features from both images using the brute-force matcher function in OpenCV (BFMatcher). Next, it finds the homography matrix and uses it to create a new image that represents the unaligned image being aligned to the reference image.
This model is a deterministic computer vision algorithm implementation and has no associated training.
Validation of the model output was done empirically, using several diverse test scenes.
The input(s) to this model must adhere to the following specifications:
This model will output the following:
See how quickly you can deploy and run models, connect to pipelines, autoscale resources, and integrate into workflows with Modzy—the ModelOps and MLOps platform
