The intelligent software toolkit for production of news TV-programs discrete function and automatons

Based on the structuring of business processes of the TV and Radio Company "GTRK Ural", automation tools have been developed that have significantly increased the efficiency of the production of news programs. The designer of interfaces was developed and introduced, due to what the problem of necessity of repeated data input is solved. The constructor is a JavaScript file that is called directly from Adobe After Effects and works in two stages: the user constructs input fields and interface controls, then the Adobe After Effects form and project are linked through the control layer, saving to JSX files and further use. The management tools of the already prepared media products have been implemented. So, for the development and filling of computer graphics templates in the preparation of weather forecast videos, automation technology was used based on the expressions and scenarios of the Adobe After Effects program and the toolkit for overlaying graphics on the video image was created. The weather forecast generator module generates a common weather data file in XML format based on hydrometeorological information. The file, convenient for machine reading, contains all the necessary data for the preparation of weather information posters during the day. Through the ergonomic interface, data are entered once, without duplication, with the ability to edit existing, including archived, data. At the same time, the chain of business processes instead of six actions before automation was reduced to two actions: obtaining a Word document from the hydrometeorological center and entering this data into the developed module, which takes about 20 minutes a day instead of the previous 4 hours. The converter for ready-made clips has been created to bring sound and video to unified standards, working through a special interface written in Python. With a view to efficiently structuring and archiving video material and subsequent search in the archive, it was developed the system that allows to automate processes of recognizing the contents of video files using neural networks based on machine learning. The recognition process is performed by means of a convolutional neural network, an open Keras library was used, tuned to work with deep learning networks. Images are classified according to three models: Label Detection, a VGG16 convolutional network of 16 layers was used for recognition; the definition of a place inside or outside the room (Indoor / Outdoor); Season recognition, the model consists of 13 layers of VGG16 network feature highlighting plus 3 layers for classification. After the introduction of the system in the working process, the time spent on the production of media products before and after automation was measured for all types of production. The total time savings per day was 320 minutes: instead of 557 minutes 236 were spent, i.e., 68% time savings. The quality of materials was improved, the number of errors was reduced due to convenient data entry interfaces and the absence of duplicate processes.

Keywords

Authors

References