While deep learning can extract good features for high-level applications, there are areas that require pixel level matching to compute geometric information from an image. Some of the applications that use this information are:
- Drones: In commercial robots like drones, the image sequence is used to compute the motion of the camera mounted on them. This helps them to make robust motion estimations and, in addition to other Inertial Measurement Units (IMU) such as gyroscopes, accelerometers, and so on, the overall motion is estimated more accurately.
- Image editing applications: Smartphones and professional applications for image editing include tools like panorama creation, image stitching, and so on. These apps compute orientation from common pixels across image samples and align the images in one target orientation. The resulting image looks as if it has been stitched...