Michael Long, Segment Marketing Manager at Analog Devices, Inc. explores the emerging and increasingly evolving security industry and how processing technology is at the core of these latest developments
Over the last few years the use of video/image analysis in security applications for the verification of threats and to identify suspicious behaviour has found its way into end equipment of varying cost and complexity. Scaling from intelligent motion sensors included in residential intrusion detection systems up to megapixel class, HD surveillance cameras; the use of video analysis through the deployment of programmable DSP solutions is becoming more and more pervasive. Currently there are processors available on the market that offer a wide range of performance options at low power consumption levels enabling end equipment solutions applicable to multiple deployment models within the industry today.
Traditional video surveillance systems deployed within municipal, government, university, and commercial buildings are undergoing a significant evolution moving from analogue cameras and centralised video processing to intelligent networked cameras which contain most, if not all, of the required image processing functionality out at the edge of the system. This shift enables not only the increase in image resolution from traditional standard definition (720 x 480 pixels) to high definition and beyond (720p or greater) but also the inclusion of real-time image analysis covering a multitude of behaviours and activities.
Each variety of deployment has focus behaviours it tracks and monitors in order to provide immediate verification of an alarm or violation condition. Examples of such behaviours include loiter detection, restricted area breech, left object detection, and suspect identification. These analyses are typically performed on a down-scaled video stream, operating at resolutions below the high definition capture level but nonetheless are intensive in terms of processor horsepower and memory bandwidth.
Each function has a task list beginning with the separation of foreground items of interest from the background comprised of either static or unrelated elements. Once the items of interest are identified, an edge map of these items must be created to allow for the verification and classifications of the objects. These functions are extensive math operations made up of several compute intensive tasks which prove challenging for traditional off-the-shelf processors such as general purpose DSPs and FPGAs.
Recognising the need to reduce the cost, power, and overall budget of real-time analytic processing, Analogue Devices has recently introduced a new family of dual core processors (ADSP-BF60x) which delivers the capability required to the support multiple behaviour analyses in parallel required by higher end IP cameras in the traditional video surveillance portion of the security market.
The family of devices includes pipelined acceleration elements which off-load some of the most compute intensive functions from the processor core and take advantage of local memory and streamlined data movement to eliminate a great deal of the memory accesses typically required when performing video analytics. As the macro-trend of real-time video analyses continues to increase in adoption, whether simply to enhance storage efficiency or to truly intercept violators at work, the value proposition of acceleration based solutions such as the Blackfin ADSP-BF60x series will map to an ever increasing portion of the security market.
On the lower end of market, there is an emerging trend to add entry level video/image verification within the typical motion sensors found within the numerous versions of residential and commercial intrusion detection systems.
Traditional intrusion detection systems include motion sensors which utilise passive infrared (PIR) technology to determine the presence of threats in the field of view. This PIR technology is effective in capturing movement but does not allow for any level of stimulus verification. The need to validate a perceived alarm condition within commercial and residential buildings is becoming a mandate in many regions of the world. Specifically in Latin and South American countries, local and national government are beginning to require the evidence of a true threat prior to dispatching emergency personnel to field locations.
This trend and resulting mandate has left equipment manufacturers and system integrators searching for a solution which will satisfy these requirements without severely impacting system cost and complexity. The resulting solution has been new ‘intelligent’ motion sensors entering the market delivering dormant image capture or analysis elements which are activated when the PIR element has been tripped.
In the first generation of solutions, compressed image or video data would be captured and transmitted to central monitoring stations as verification of threat stimulus. While many systems will continue to use this method, there is a push to eliminate the need to compress, store, and transfer actually image or video data through the use of video analytics.
In this scenario, the image data is analysed to classify the detected objects in a similar fashion to the behaviour analysis functionality outlined in the surveillance camera application above, however the intelligent motion sensor use case does not require behaviour verification but rather simply a classification of the motion source. By virtue of this lower level of analysis, the task can be performed exclusively in the programmable domain by a more cost and power consumption conscious solution.
The company’s entry level ADSP-BF51x Blackfin processor enables both wired and wireless ‘intelligent’ sensors to be included across the full range of intrusion detection system deployments. The advent of this technology delivers absolute threat verification at acceptable price points without sacrificing critical system criteria such as ease of use and maintenance-free utility (most commonly manifested as elongated battery life).
As security applications gain in widespread adoption and the underlying analysis tasks make their way into adjacent applications such as lighting and HVAC occupancy detectors, Devices such as the processors described here will continue to offer a very compelling technology solution for end equipment adoption and differentiation.