Application of neural networks using machine learning for live video recognition

DATA COLLECTION

We analyze and process in real time a live broadcast from one of the public cameras at Gdansk International Airport in Poland.

At intervals of 10 seconds we obtain a frame of the emission we process, we cut it in the area defined by the green rectangle of the video player, since that is the area we are interested in analyzing and we manage to reduce the amount of information to process. We store and analyze the resulting image with different automatic learning models to obtain the necessary information.

DATA PROCESSING

In the analysis phase we carry out two tasks, a real time analysis applying the models to obtain the data shown on the web (without storing them) and which identify separately whether or not there is a parked aircraft, in addition to obtaining the airline, the aircraft model and whether or not there are vehicles for fuel loading and baggage management.

We carry out these analyses using two different recognition systems, on the one hand we apply automatic learning models programmed and trained by us (our own models and scripts) and in parallel we carry out the same task using Cloud Auto ML (a package of products that makes it possible to create customised models easily). With this double processing we increase the reliability of the results by comparison, and we improve the training processes of our own models and scripts.

In addition to real time processing, we store all collected images for a complete overnight reprocessing. This is useful because depending on the project to which these technologies are applied, it will not always be necessary to obtain data in real time, and in this case, the data processed in batches are stored in Big Query for later analysis obtaining different indicators.

DATA MINING

With the previously stored data we obtain different indicators and statistical data such as average parking times per company, refuelling times, number of daily flights parked, etc … which allows us to show an example of the exploitation of data that could be done with the analysis made on the video.