Orthogonal Mosaic principles for Data Analysis

In this brief article Inbright Image Acquisition Team will explain basic concepts on image reconstruction for an effective orthogonal mosaic, have this in mind when trying to analyze data obtained from a Drone or Satellite.

One of the first problems encountered when trying to obtain data to analyze after with computer vision algorithms, is How to acquire consistent data from a camera mounted on a Drone?

Drones move at a speed that vary from 3 m/s to 10 m/s for Multi-rotor and even if you are able to automate movement to make the Drone stay in a single position, there are several factors that will make almost impossible to guarantee you will take a picture in the same position to have a correct mosaic constructed, to name a few:


Air is one of the main environmental factors that will always be a problem when piloting a drone, even if this is in automated pilot by software, this will not guarantee that the Drone will stay in one place, this is something you cannot control and the flight card will never respond so quick that air will not move your device.


To put it this simple, hardware is not precise, Flight ESC´s wont deliver the precise Voltage in the same way, same time for every motor; even though, I think this are one of the least to worry about, if proper maintenance and change of Hardware pieces are done.


Depending on the brand, precision and time of the day (HDOP) GPS can hurt you with centimeters precision or meters, you cannot really trust it at all, period.


Most of the reasons things fail when a Drone is acquiring images have to do with you, knowing your equipment, having a proper calibration of the systems, correct set up of Gimbal and even the slightness thing can change the whole play, for the therms of this article we are considering you already know how to do all this.

Now, hands on matter we are going to talk about some of the things to consider to construct an effective Mosaic Having the following information from each image will help reconstruct the Mosaic in an organized and consistent way, this information may be used to identify and take actions to correct image if necessary.

XYZ Position


Since as we already said, the Drone is not going to be at the same point within all the photos, knowing the precise position at the moment the picture was taken will help taking the right decisions to have two images at two different XYZ positions represent the same area. Usually XY is called Overlap and it is needed to recognize the borders within images in order to construct a consistent frame, by the other hand Z is treated as a pixel representation portion, in a picture of a plant at 10 meters a pixel will not represent the same millimeters as a picture of the same plant at 5 meters, therefore all this information is needed to create a normalized mosaic where at any pixel in the given mosaic, it will always represent the same known distance on real life.


Axis Rotation

Axis rotation is usually covered with the Gimbal, if properly set up you will not have a problem with this, tough it may be useful to save and double check every single photo to track any misbehavior with Gimbal. Not having a Gimbal represents other kinds of problems such as perspective, that topic requires other approach which is not covered in this article.

Camera Sensor

Camera_Sensor Knowing the resolution and pixel size from the Camera sensor you are using will help to determine from each picture the conversion between light taken from sensor to pixel size in a real representation.

Relative North


This is a very important factor to know where is the drone pointing when taking the picture, if the Drone has moved slightly any degree Right-Left, this will have huge impact for a proper stitching of the picture, imagine you are solving a puzzle and you try to put a given piece in the exact same position you took it from the table, things need to fit exactly to have a proper image, either you know the rotation relative to a given point (usually north works) or else you may find yourself with something like the image below.


With all this now you know all that implies creating a Mosaic to analyze after with Computer Vision Algorithms, this were just simple principles, the real implementation of all this is a very precise and complex process you should execute very carefully.

If you have any questions or comments feel free to email us to Hi Inbright

Sincerely, The Inbright Image Acquisition Team.

This article is my 12th oldest. It is 790 words long