Video Analytics for automatic videosurveillance

VTrack technology is the Video Analytics platform created by TechnoAware for the development of automatic video surveillance products and solutions.

Established from the experience gained by the University of Genoa’s ISIP40 Group over more than 25 years of research activity, VTrack technology collects in software libraries the most technologically advanced algorithms and methods for the automatic and real time video flows analysis.

Using standard video cameras and PC architectures, thanks to VTrack technology it is possible to detect, track and classify subjects (individuals, vehicles, objects, …) or events of interest, for the below types of application:

Public security Stations, airports, museums, places of worship, public buildings, urban areas, defence, … Private security Residential areas, apartments, car-parks, industrial areas, … Traffic management and Info-mobility Vehicular flow analysis, average speeds, queues, accidents, forbidden ways or transits, … Marketing and Retail Monitoring of customer flows, control of sensitive areas, department flows, behaviour analysis, … Domotics Access control, monitoring of elderly person, virtual guide, … Environmental monitoring Monitoring of natural areas, fire detection, monitoring of animal flows, … Entertainment Analysis of sporting events, interactive games, …

Written in C++ software language, both for Windows and Linux O.S., VTrack modules has been made to be integrated in any third party product, solution of platform.

Thanks to VTrack technology and to the expertise of TechnoAware’s team it is possible to develop products for standard functionalities or to custom ad hoc modules for specific applications.

VTrack benefits

• Flexible, modular software platform developed in C++ language and available for both Windows and Linux O.S.
• Can be integrated with any PC-based platform or architecture
• Compatible with analogue, digital, IP video cameras and offline video flows
• Automatic, real-time alerts for only events of interest
• Robust and reliable, able to reduce efficiently false alarms due to light conditions, dynamic backgrounds, atmospheric phenomena and subjects that are not of interest, …
• Continuous level of attention
• Enhanced privacy protection, an electronic processor being, for the most part, the only spectator of events occurring in the scene
• Ad hoc module customisation option for certain applications

Technical requirements

• Acquisition of video flows from analogue cameras via DirectX or VideoForLinux compatible video grabbers
• Acquisition of video flows from compatible IP cameras or video servers (Milestone, Axis, Acti, Lumenera)
• Acquisition of compressed video flows in all standard formats (MJPEG, MPEG4, H264, …)
• Minimum area for the subject of interest to be detected with maximum efficacy: 50-70pixel
• Minimum frame rate for optimal performance: 5-7fps*
• Onere computazionale**:
  
  * No such performance improvement seen for frame rate higher than 10-12fps
  ** Values based on experimental tests for a module of average complexity

Methodologies and algorithms

VTrack technology can be likened to a large library, where it is collected and continuously updated all image processing state of the art for the automatic and real time video processing of scenes. This is why VTrack, unlike traditional motion detection based modules, can guarantee unprecedented functionalities and performance.

Below are some examples of the algorithms and methods used:

Low level algorithms (signal, feature): IP and analogue video flows acquisition, cameras calibration (f.e. by Tsai’s method)), linear and non-linear filtering algorithms, sub-images extraction, Hough Transform for lines, ellipsis, circles and generalized, corners and edges extraction, change detection simple and non-stationary background based, foreground segmentation by color and/or gradients statistics, dynamic background upgrading (f.e. by Gaussian Mixtures), morphology, resampling, connected components, color spaces conversion, image stabilization, shadows and ghosts detection and removal, color segmentation, gamma correction, contour enhancement, …

Medium level algorithms (focus of attention, classification, tracking): supervisioned and non-supervisioned classifiers, k-medie, fuzzy c-medie, shoslif, support vector machines, multi-layer perceptron, self-organizing maps, … Moving subjects tracking by bounding box overlapping algorithms, mean shift, condensation, corners tracking (nlse tracker), trajectories, crowd analysis, klt tracking, hidden markov model, graphs, minimum spanning trees, kalman filter, particle filter, multiple gaussian decomposition, …

High level algorithms (events): events of interest modelling, behaviour analysis, events analysis, epecific functionalities