Summary

Video analysis consists in providing to computers the capability of understanding the information contained in the video sequences. There are many real-time applications for video understanding, as behaviour recognition, access and traffic control, elderly care at distance, automatic patient supervision, human/animal/crowd behaviour analysis, among many others.

This software is a general behaviour analysis platform able to extract and learn valuable information from noisy video scenes for realtime applications. The main contribution expected for this platform is its capability of learning meaningful behaviours of objects present in the scene, even in presence of noisy information. The learning process of these behaviours can be guided to obtain meaningful information by estimating the reliability of the attribute information associated to the object models and processes. The estimation of reliability measures allows to properly model the uncertainty of data due to noisy videos and multiple other difficulties present in video applications (e.g. illumination changes, object occlusion, high volume of data).

An important objective of this new platform is to treat the video understanding problem in a generic way. This implies implementing a platform able to classify and track diverse objects, to work on different scene and video configurations, and to easily adapt the utilised algorithms. This generality will allow to quickly adapt the platform to different applications, and to serve as basis for future research projects on even more complex situations.

The Software

Version

3.0

License

GNU GPL v3

More about this Project: