The drones, together with the powerful cameras they equip, are starting to become tools that go far beyond recording videos for leisure and work. There are companies that are already using them to analyze the traffic of vehicles and pedestrians. These companies are taking advantage of the research and data, even to be able to manage parking spaces and raise the detection of bad drivers in real-time.
Nvidia spoke with DataFromSky, a Czech startup born in Brno in 2013 that extracts and analyzes data from the videos that a series of drones record from a certain road area. The applications that there are for this type of analysis are many more than what is usually believed.
Let’s start by reviewing exactly what DataFromSky does with its drones. It is a process in two phases: first, they use those drones to record the traffic of people or vehicles, and then they use their analysis software to analyze the resulting videos and extract the data that we see captured in the video that you can see above. The analysis is done later from the offices of DataFromSky.
These data can range from vehicle acceleration and/or deceleration patterns in certain complicated traffic points such as roundabouts or crossings to the vehicle count itself, determining traffic flow or density. They can be used to decide how to improve that traffic zone, or how to limit the traffic if in some sections the vehicles accelerate too much. At the moment, they do it in Norway, Sweden, Denmark, Finland, the Netherlands, the United Arab Emirates (Dubai), Mexico, the United States, Australia and Belgium among others.
How do they get those videos? Engineer Lenka Šedivá explains how drones’ deployment depends on each country’s regulation. The objective is to cover a certain traffic area, with each drone covering a surface “block” of approximately 850 meters maximum length. Still, some countries have to cover that distance with more drones due to the legislation.
The recommended elevation for making the recordings is around 100-150 meters so that each vehicle “occupies” an average of 16 pixels in the resulting video and thus can be well analyzed. The total verticality is the best angle to make the recording, whenever possible. Always avoid any kind of reflection that the sun can do, and if for example, the wind is very strong that day you can stabilize the video after recording it through editing software. It is not always recorded in these ideal conditions: the laws of some countries such as Denmark or Singapore require recording from 80 or 60 meters respectively.
David Herman, CEO of DataFromSky, tells us that the cameras included in drones like those of DJI are more than enough to make the recordings with the necessary quality. In specific cases such as nighttime recordings (lower video) or with a fish-eye camera, night filters or calibrations can be used in the analyzes of the videos to avoid too pronounced distortions.
The executive does not give us too many details about the algorithm used to detect all bodies in movement and generate data. Still, he confirms that his development began at the University of Technology in Brno in 2013, collaborating with that same institution.
By testing several modes of analysis using “large data sets” taken from their videos they have obtained their current software, which they sell commercially to any company that is interested. They only manually enter the points as crossings of streets and roundabouts, defining those in which you have to record the speed data or the number of objects that pass through there. And it is something that they are already working to automate.