Law enforcement agencies are under mounting pressure to provide a safe and secure environment for their citizens. Cities are increasingly using video surveillance as a tool to fight crime, prevent terrorism, and increase the personal safety of citizens. Visual surveillance for urban surveillance is currently a highly active research area in image processing and pattern recognition. The number of studies published in the last 10 years out numbers all previous related literature three-fold.

Behavior analysis algorithms rely heavily on the core technology available. There are many limiting factors to the usability of these core technologies in real urban surveillance environment. Implementing analytic on some videos may not be feasible or could be restricted to only a subset of the algorithms available. There are many hardware-related problems such as poor resolution, low frame-rates, or insufficient processing hardware. For instance, crowd monitoring algorithms usually rely on the calculation of optical flow, which requires a moderately high frame-rate and significant processing power.

Most requirements today are around digital IP cameras, high megapixel resolution imaging. However the limitation of centralized processing push operators and integrators to down scale the megapixel image to 4CIF some times to 1CIF. In fact if you have 300 units of 3megapixel cameras streaming at 20fps you will need super machines to do any type of computation. Of course another factor is the recording cost So why megapixel IP cameras then?

Most vendors today limit their offering to the following applications: Object tracking, Breach (trip wire), loitering, crowd analysis, stance change, object left, object removed. The following table highlight the main vendors marketing those applications on their web sites.

I can see those classes of analytic working very well for infrastructure protection in a controlled environment that what companies like Nice systems learned very early and were successful when they focused on specific markets to target specific environments. Urban surveillance and intelligent transportation however offer different challenges and complexity. The uncontrolled nature of the environment, unpredictable behaviors require advanced algorithms and techniques requiring huge computational power causing a scalability problem.

Video analytic using distributed AVISTA smart sensor networks are emerging today as an alternative to CCTV and existing video analytic technologies, both in the field of surveillance (airports, train stations, museums, public spots) and intelligent transportation.

Aligning operational requirements and building winning strategies using existing technologies and cutting edge technology and most focusing on ROI will be the key today in building leadership in this area. The vendor driven video analytic strategies proved not working and today a clear understanding is built around customer driven video analytic and services.

No doubt that traditional centralized approaches offer several drawbacks, due to limited communication bandwidth, computational requirements and thus also limited spatial camera resolution and frame rate.

Some integrators/solution providers start moving the processing to the edge, and putting a PCB DSP boards at the pole where the camera is installed�.it is a natural evolution and eventually intelligence will move to the edge.

A network-enabled AVISTA Smart Imaging Sensors for real time distributed intelligence is capable of tracking objects and extracts key video features in real-time and offers a very bandwidth-conservative approach, as it only transmits XML meta data results of the On- sensor processing which are of a higher level of abstraction.

Conventional computer vision systems typically see cameras only as simple damp CAMERAS. The processing is performed after transmitting the complete raw video stream via a costly and often distance-limited connection to a centralized processing unit (SERVER).

The AVISTA smart imaging sensors are built on the concept that it is more natural to physically embody the processing in the imaging sensor itself: what algorithmically belongs to the camera is also physically performed in the camera.
The idea is to compute the raw sensory data where it becomes available � directly at the sensor � The computation is pure feature extraction and tracking parameters.

Once the processing performed transmit only results in a format of XML metadata that are on a higher level of abstraction. This follows the emerging trend of self contained and networking capable smart sensors.

We were able to deliver some outstanding solutions as part of our DST analytic library in the last few months. The latest functionality was near miss pedestrian/vehicle detection at an intersection. Transportation agencies are looking to enhance pedestrian safety by enhancing traffic light systems. The data we provide to our clients is around traffic flow and specific incidents. The best article describing the problem appeared in the NY time recently.
Recently DigiSensory has been engaged by RTA in NSW to deliver the near miss detection technology. The following video show case th performance of the system at an intersection in Wolongong (NSW):

From video algorithms to business intelligence: the Avista Smart Agent is a powerful web based platform allowing real time traffic monitoring and incident detection.

Furthermore the Avista Smart Agent (ASA) correlates information and intelligence between different AVISTA smart sensors and provide the end user and operator a high level of intelligence.
In real time over a web based application platform that can easily integrate intelligence visually on google map or any GIS mapping technology. The user can access the real time sensory driven intelligence, traffic flow and will be notified of any incident.


For RTA and department of transportation around the world a simple application like near miss detection or crash detection will allow the traffic management operator to plan better and build more advanced system to reduce fatalities and incidents at intersections. More applications related to smart intersections are being build with partnership with different developers and R&D institutions. By offering the ASA on a open platform we are working towards building a community of AVISTA application developers working towards building speciffic rules and applications to meet different business objectives. While the image processing core is built at the sensor level the ASA server use simple XML programing to define speciffic rules to build different applications.



The following images demonstrate also the easy access to any data by a structured video/image indexing. The operator will be able to access any video/imaging data just by event ID. In a city like London with 4 million CCTV (1 camera per 15 citizens) video search will take hours and hours of intensive labor.

The Avista smart imaging sensors platform is more then what conventional CCTV, IP cameras, and analytic can offer. It is a management intelligence driven platform and a business operation tool where visual data is protected and processed by advanced image processing technique at the sensor level coding all visual information to XML meta data. The following video example shows visual sensory meta data information sent from the Avista smart sensor and the ASA predicts an assault.