DEEPFRAMEWORK
Distributed and scalable Framework for analyzing a real-time video stream
Contact
Challenge
The analysis of a scene captured by a camera can represent a problem both in terms of cost and reliability as the amount of information to be extracted from the video increases. If this analysis were carried out by a human operator, it is evident that it would be difficult, for example, to manually count a certain type of object and describe its characteristics in real time. The solution adopted to solve this type of problem is provided by the study of computer vision techniques. In recent years, the techniques of Deep Learning have been particularly successful thanks to their reliability and to the ever-increasing power of calculation provided, at relatively contained costs, by the hardware supports. The criticality in the use of these techniques is represented by the execution time of the algorithm. It is clear that, even though accurate classifications are obtained, as the number of information that needs to be calculated increases, the response time of the algorithm will grow if the appropriate solutions are not adopted.
Overview
The DEEP-Framework is a Python-based distributed and scalable framework for analyzing a real-time video stream. At its core, the framework provides a modular Docker-based pipeline that allows to distribute and parallelize all tasks from video capturing, to object detection, to information extraction, to results collection, to output streaming.
The current version includes an implementation of following pipelines:
- Age estimation;
- Gender estimation;
- Face recognition;
- Glasses detection;
- Yaw estimation;
- Pitch estimation.
The Deep Framework also provides the ability to develop and include your own detection or feature extraction algorithm in the execution pipeline quickly and easily.
Innovative features
- Multiple Deep Learning algorithms can be executed in real time;
- Allows to develop a computer vision algorithm and include it in an execution pipeline quickly and easily.
Potential users
- System developers of:
- Computer vision algorithms;
- Augmented Reality;
- Presentation of personalised contents;
- Telemonitoring and video surveillance
- Security operators, retail.
Impact sectors
Security, intelligent environments, business analytics, marketing, entertainment, retail.
Other resources