Smart Adapter for IP Cameras
keywords DEEP LEARNING, VIDEO ANALYSIS
Reference persons ENRICO MAGLI
External reference persons Ferdinando Ricchiuti (CSP)
Thesis type APPLIED RESEARCH
Description In the last few years, smart cameras are emerging in product lines of the most important vendors. These devices are capable to implement local visual processing in order to detect specific objects or situations. The focus on this kind of devices is limited to applications related to security and traffic monitoring.
Even if some camera devices comes with their own development environment (allowing developers to enrich their functionalities) it is not easy to extend detection capability to address other kind of applications.
The objective of the work is to design and implement and hardware and software adapter, that can be connected (via IP) with a standard IP camera. The device can be configured with a trained Neural Network and will do auto tagging of objects recognized inside the frames captured by the camera. This annotation will be made available using REST interface to allow the end-user application, any kind of further processing. The trained Neural Network should be provided using standard formats, like DMG’s Predictive Model Markup Language (PMML).
Required skills Candidate students should have a background in C/C++ programming and Internet protocols. Knowledge of neural networks (including TensorFlow environment), OpenCV, Python and Linux programming (in particular on embedded systems) are not prerequisites, although they would help in the initial stage of the activity.
Notes This thesis will be performed at CSP (Centro Supercalcolo Piemonte)
Deadline 15/03/2019 PROPONI LA TUA CANDIDATURA