The detection of birds
The software, according to TEC, receives a “video feed” (a filmed sequence) that covers a radius of over two kilometers around an airport, activates a system for processing data in real time and detects the position and species of all birds in the area, automatically triggering the appropriate warning calls to completely disperse the flock.
“Thanks to an interdisciplinary team of biology and technology experts – adds Masci -, we have trained the neural network at the heart of the software by analyzing over 100 thousand images of different species of birds and today the solution is able to recognize birds for number, species and position with an accuracy close to 100%, a level much higher than that achievable by a human observer. The data we acquire allows us to cover the entire critical area and to refine the use of specific deterrents with extreme accuracy: we cannot guarantee the elimination of the risk of accidents but we certainly guarantee greater prevention and a reduction of the phenomenon ».
During the development of the software and for the first tests at the Verona Villafranca airport, Tec had opted for a distributed IT platform, which included desktop PCs dedicated to video processing and units server rack separate for data management and storage. Even before launching BCMS Ventur on the market, the decision was made to change in the running so as not to jeopardize the implementation of the solution in terms of reliability, cost efficiency and manageability of the basic IT infrastructure.
The intuition to lean on a appliance “All-in-one” to physically implement the solution has found support in Lenovo, which has tailored a fully scalable architecture for the startup (as well as remotely executable for standard management and maintenance activities) and based on a ThinkSystem SR650 AI server with Nvidia Quadro RTX 4000 GPU.
The High Performance Computing system in question, as also assured by Valerio Rizzo, Lenovo’s Emea AI Solution Architect, uses highly resilient virtual machines, always active in 24×7 mode, and with a single unit it allows to analyze the video streams in 4K resolution coming from from up to 18 cameras simultaneously.