The abbreviation " VPN " has slowly become established among average Internet users, which is certainly good, but the problem is that people have started to see VPN as the ultimate solution to all problems.
If you've ever used Street View, you've probably noticed some irregularities in the form of truncated objects. These irregularities are due to the poor merging of images produced by Google Mapping Car. This problem is solved now and you will notice the results soon.
What is it all about?
Google maps the streets (and various areas of the world) using a special set of cameras located in one sphere, and all this is called "Rosette". That sphere is located on top of cars, backpacks, camels (yes, and camels are used for mapping), boats, etc. As the car moves, the camera takes a set of pictures. Images are stored, sent, processed, "glued" to an expanded sphere, and finally, a sphere is created that users can see.
In the whole process, the biggest problem is to fit the images so that they are aligned without cut parts. Because, these images from all the cameras overlap at least a little clean so that there would be no gaps, and so that the algorithm (old) would make the transition as well as possible. In the pictures below you can see what this overlap looks like (old algorithm).
Since there are many pictures, it is impossible to correct these parts (ie it is too expensive to pay so many workers), so an automated solution was sought. And the solution is a new algorithm that visually fits the layers of images so that they match perfectly.
This algorithm detects lines and shapes in images and tries to match them. After "dragging" them to the ideal position, he must then fix the rest of the panorama so that the other parts do not "unravel". Repairs are done only after all visual defects have been resolved.
In the example below you can see how the algorithm aligns two layers.
After all this process, a clear difference can be seen. In the pictures below you can see before and after the example of the bridge.
The biggest problem is if this defect in the picture happens on a logo, title, or name… because significant parts can be lost. In the following image, you can see how the algorithm corrected all the irregularities on the parts with the text, as well as on the circular contours.
A video of the presentation of the new algorithm states that Street View images will be fixed all over the world very quickly, so check back from time to time.
Comments
Post a Comment