DataFromSky is an advanced tool designed for monitoring and analyzing traffic flow in road networks by processing video data.

Thanks to long-term development of specialized algorithms for detection and tracking, DataFromSky is able to provide comprehensive information about the behavior of road users solely from aerial video data taken by an ordinary camera.
The camera may be carried by an UAV or mounted on a very tall building. DataFromSky enables fast and accurate surveys of road networks. In addition, it can be easily used in a wide spectrum of many other different applications.

Convert a low-cost UAV into
a very agile mobile sensing platform

Unlike in many commonly used systems for traffic monitoring, the only requirements for detailed analysis of road user behavior by DataFromSky tool is an aerial video and a description of the recorded scene.

The video may be taken by a UAV, or a camera mounted on a mast. Our detection algorithms are capable of recognizing vehicles as small as 12×12 pixels to allow for coverage of a large area by one device only (more than 30,000 m2).

As high-quality video stabilization algorithms are an integral part of DataFromSky tool, the sensing platform is not required to be 100% stable during the recording.

The ultimate solution
for traffic analysis

DataFromSky is a unique solution thanks to comprehensive data which it provides. The output from the first phase of the analysis is a list of trajectories of the detected vehicles.

Each vehicle is labeled by a unique ID and all its positions, speeds, and accelerations are recorded during its passage through the monitored section.

Thanks to extraction of spatiotemporal information about each vehicle in the scene, DataFromSky offers unlimited possibilities in data interpretation and inspection.

Advanced analysis –
the absence of limitations

DataFromSky also includes a tool for advanced data analysis. A virtual counting gate may be defined anywhere in the input video to establish the number of vehicles in this place.

Further, you can also obtain detailed information about traffic lane utilization, sectional measurements, average speeds, observed anomalies, etc. The tool also provides advanced functions for traffic engineers such as Tg and Tf estimations, and computation of vehicle lateral acceleration.

The data may also be exported in the CSV format and further analyzed in your favorite software.

Software architecture –
behind the scenes

The overall system can be divided into three main parts: preprocessing, vehicle detection, and tracking. In the preprocessing step, the acquired image gets undistorted and geo-registered against a user-selected reference frame. The efficiency and robustness of vehicle detection methods in aerial images have been addressed several times in the past.

We designed a highly optimized multi-layer detector and trained it on a hand annotated training dataset with 10.2M positive and 10.2M negative samples. The detector is capable of learning from its errors and improving over time.

The tracking part is an extremely challenging task due to the presence of noise, occlusion, dynamic and cluttered backgrounds, and changes in appearance of the target. A subtype of sequential importance sampling filter has been employed to deal with these challenges. Due to the outstanding performance of detection and tracking algorithms, we can provide high-quality data and guarantee a hit rate of over 96%.