Presented at the 39th Annual Meeting of the Institute of Nuclear Materials Management, July 1998

ABSTRACT

Present surveillance camera systems have become overwhelmed by new remote monitoring requirements, multiple camera requirements, smart video (motion detection and object recognition), and very high processing power functions such as authentication, encryption, and very high speed communication systems. These requirements demand a system that possesses abundant processing power, as well as ease of expandability to handle a myriad of real time tasks. As a result, surveillance cameras and their video processing units must be 1) capable of very high speed communications, 2) highly adaptable from a software perspective for future requirements and a hardware perspective for increasing data throughput, 3) internally low power, and 4) extremely stable and reliable, In addition, the functions of the video processing unit should be scaleable to support both simple security operations or more complex safeguards operations in a cost-effective manner. This paper investigates how current and future surveillance systems will meet these new demands.

INTRODUCTION

The basic Safeguards requirement for verifiably uninterrupted surveillance has remained largely unchanged since its inception, and for many years the inspection scenarios and implementations that met the needs of video surveillance were essentially stable. Manually collected and reviewed time-lapse photography was the mode of operation and there was no interaction between the surveillance system and any other instrumentation. The demands on the surveillance system were very simple¾ reliable unattended operation.

While this fundamental Safeguards requirement has remained essentially the same, changes to the inspection paradigm to admit automated review and remote monitoring have dramatically increased the demands on the Safeguards surveillance systems. The technology has grown nearly exponentially in response to those demands, and in only 5 years Safeguards surveillance systems have evolved to include self-contained, cryptographically authenticating, motion-detecting cameras operating with a server that provides remote data access.

BACKGROUND

In the beginning, video surveillance was defined to be the unattended time-lapse recording of pictures. The inspectors collected the pictures approximately every 3 months and manually reviewed all of them, seeking to find recorded verification of the facility activity logs, and occasionally finding evidence of activities requiring further explanation. The surveillance systems that collected the images were very simple, but the review process was incredibly time consuming. The Twin Minolta system that collected the pictures was very simple and reliable, consisting of a pair of film cameras that were triggered with a timer and housed in a tamperproof enclosure. The review process entailed developing the film and then examining the thousands of pictures individually. The simplicity of the data collection system was "paid for" by the arduous review cycle.

The continuing need for more and more cameras compounded the workload inherent in the manual review process. The clear need for relief from the burden of the review cycle created the demand for an automated review process, or at least an assisted review process. However, the basic film photography technology simply was not consistent with an automated review concept. This demand spurred the development of the next generation of video surveillance systems which used analog video tape recording instead of photographic film. The "automated" review concept was simple¾ the pictures could be viewed on TV as a movie instead of as individual pictures. MIVS became the basic single-camera recording system, and MOS and MUX addressed the demand for multi-camera systems. However, it soon became clear that reviewing the surveillance on TV was not sufficiently automated to provide the relief necessary. This fostered the development of a family of computer-assisted review stations. The GRS and Mark IV were eventually supplanted by the far more capable MARS and MORE, which brought automated motion detection technology into the mainstream of Safeguards review.

Despite the quantum leap in review capability afforded by MARS and MORE, it was clear that the ability to provide additional review automation was restricted by the limitations of the analog recording medium. The acknowledged reliability problems of the analog recorders and the recognition of the limitations imposed by the analog media fostered the development of the first all-digital Safeguards surveillance system, GEMINI. The digital camera also provided the first opportunity to satisfy another long-standing need: cryptographic authentication in the camera housing itself. Perhaps more significant is the review capability that derived from the all-digital environment. The General Advanced Review Station, GARS, is the first surveillance review station that does not require specialized hardware, and since the surveillance data arrives totally in digital format, the GARS is a pure software product. More importantly, GARS provides the many review functions and features that were simply unobtainable with the earlier technology, including automated back-end motion detection, automated report generation, automated anomaly detection, picture enhancement, scrolling, jogging and many others.

Up until this time the surveillance paradigm remained unchanged¾ time-lapse photography with data recovered manually by an inspector at the site. However, at about this time, increasing emphasis began to be placed on reducing the volume of data to be reviewed and on the concept of remote monitoring as a means of streamlining the entire inspection process.

RECENT PROGRESS

It was immediately clear that standard surveillance records were simply too massive to routinely transfer electronically within any conceivable budgets. It was equally clear that a credible Front-end Motion Detection (FEMD) technology could solve this problem by recording only those pictures that showed changes during the surveillance. Significant development and testing effort was invested to meet the demand for FEMD, and the first actual remote monitoring system was installed in a storage vault in South Africa. This system used GEMINI with the motion detection processing being performed in the digital camera itself and a Windows NT server to collect the surveillance data from the GEMINI and to handle the remote data transfer to Vienna. The successful operation of that system permanently validated the concept of remote monitoring, but the quality of the imagery was not completely satisfactory. The blurred images resulted from using a snapshot camera to capture images of objects in motion¾ a well known problem in film photography. At about this time, the VDIS camera by Dr. Neumann GmbH had completed testing and was approved for Safeguards use. Rather than being adapted for motion detection, the VDIS was designed from inception to be a motion detecting camera. VDIS solved the blurring problem inherent in the GEMINI camera and has proven to be a very robust answer to the demand for FEMD surveillance.

The demand for a reliable remote monitoring implementation of VDIS resulted in a collaboration between Neumann GmbH and Aquila to produce the dual-camera remote monitoring configuration that is in use in Japan. Linking Dr. Neumann's VDIS cameras to Aquila's Remote Monitoring Communication Server with cooperative software to collect, distribute, authenticate, and multi-level encrypt the alarm images has resulted in an affordable system with all of the features necessary for secure remote monitoring. Further collaborations are under way to produce a direct replacement for MIVS and a standardized multi-camera surveillance system for remote monitoring using the technologies from both companies.

At about the same time, Aquila introduced a system that brings remote monitoring capabilities to existing MOS and MUX multi-camera analog surveillance systems. The MOS/MUX Remote Monitoring System captures the video from the eight existing analog cameras, performs motion detection at the rate of one picture per second per camera, and provides the same multi-level encrypted directory structure on the server as is used for other remote monitoring applications.

As these new technologies were introduced and deployed, the field experiences indicated areas for improvement, as well as stimulated new concepts for potential solutions to problems that had always been considered "too hard."

NEW DEMANDS

Current motion detection methods are a significant improvement over the state-of-the-art methods from 3 years ago. Nonetheless, they are subject to false alarms from lighting changes. Consequently there is a new demand for improved video processing that can discriminate between generalized scene changes and true motion of objects. At the same time, there is a demand for faster detection of the onset of motion. Thus we have a demand for more complex processing to provide greater discrimination, accompanied by the demand to do the processing faster.

The realization that a significant improvement in tamper resistance can be provided by multiple camera systems organized so that each camera has another camera somewhere in its field of view has spurred the demand for multiple camera systems having advanced motion detection capabilities. Since any attempt to tamper with a camera requires some kind of motion to occur around that camera, any tampering activities will be detected and recorded by a different motion detecting camera. The demand for multiple camera systems is accompanied by a concomitant demand for lower prices so that the multiple camera systems are affordable.

Power outages are commonplace in nuclear facilities, yet the need for continuity of surveillance persists. Consequently, there is an ongoing demand for longer and longer periods of operation on battery backup. Because battery technology is quite stable, no significant advances in power density are expected anytime soon. Therefore, meeting this demand requires application of the latest very low power components from the cellular telephone and laptop industries.

NDA and other instruments, which had always been operated by the inspector in "attended" mode, are now being considered for inclusion in remote monitoring scenarios. This scenario certainly introduces the issue of data authentication for these instruments as discussed elsewhere. While the SafeComm dongle provides a mechanism for cryptographically authenticating and delivering the sensor data to the server, the question of tamperproofing still remains. The primary tamper scenario is one of shielding the detectors or substituting materials during measurements. The protection concept is to deploy motion detecting cameras to monitor the instruments, thereby capturing pictures of any activity around the instruments. It is anticipated that the sensors will alert the cameras to take pictures any time a reading exceeds some threshold; and likewise, the cameras will alert the sensors to record data anytime there is motion in the field of view. The result is an interactive monitoring system with each sensor type triggering the other sensor types so that the environment is being monitored and recorded. This scenario introduces new demands for trigger generation and response in the cameras, as well as for advanced motion detection. It is likely that these surveillance systems will need at least a modest capability of object recognition so as not to trigger the interacting instruments during routine operations.

Safeguards inspection of fuel transfers and similar operations has always been problematic. These processes involve the complex movement of material in the form of bundles, fuel rods, or other containers in the presence of moving people and machinery. The only solution to date has been for an inspector to be physically present to observe the operations. Standard time-lapse surveillance proved unsuitable in these situations. Very short intervals were needed in order to verify the direction of motion of the material with complete certainty. However, the short interval produced huge quantities of redundant pictures that were very time consuming to review. Current FEMD cameras certainly eliminate the huge volume of excess static pictures, but because they trigger on the normal incidental activity associated with the process (as they should); the result is still a large quantity of pictures having no Safeguards significance. This scenario places a new demand on the surveillance systems to recognize and record certain object shapes.

NEW DEVELOPMENTS

The technology itself is placing demands on the surveillance systems that surpass those of the apparent needs of Safeguards. For example, Safeguards normally requires pictures to be saved at intervals ranging from many seconds to a few minutes. However, Safeguards also requires that the first picture be saved within a few seconds from the onset of motion. In addition, advanced motion detection and object recognition technologies require a much faster sampling rate just to enable the algorithms to operate correctly, even though pictures will never actually be stored at the fast rate.

The recent I²SIP established an instrumentation standard that prescribes that each sensor (camera) provide authentication, local buffer storage, and Ethernet communications. Digital Safeguards cameras (e.g. GEMINI, VDIS) have included authentication and local storage since inception. Although Ethernet is an ideal network medium for transmitting pictures, none of the available cameras are currently Ethernet-capable.

The following table lists several of the more significant demands on surveillance cameras along with the features they imply and the impacts they impose.

The combination of these requirements indicates that the new generation of surveillance cameras must have a very fast video processor, large amounts of memory, Ethernet with TCP/IP, and very advanced motion detection processes, and all of this must consume nearly no power.

To begin to approach meeting these new demands, Neumann GmbH and Aquila are collaborating to integrate the SafeComm dongle inside the camera housing of the VDIS camera. This will bring Ethernet connectivity to the VDIS and make it fully I²SIP compliant. This combination will support current needs in the short term.

For perspective, it is not at all uncommon for a Pentium workstation to deliver live video (30 fps) to a screen for viewing. However, this is somewhat misleading as it is accomplished by totally bypassing the processor and memory and delivering the digitized video directly to the SVGA display controller over the PCI bus. When it is necessary for the video to be digitized and placed in memory, processed, and then compressed and stored, a 200 Mhz Pentium is entirely consumed with a single channel of live video data - and cannot authenticate the data at this rate. We know that advanced motion detection requires 4 to 6 times the processor service per picture than required by simple change-detection and that object recognition requires even more. Counteracting the increased workload is the 75% reduction in frame rate from that of live video. Combining these factors implies that the single camera processor must have a computing capacity similar to that of a 200 Mhz Pentium!

In the longer term, the new generation of processors from the cellular phone industry offer the potential of providing the computational capacity needed for advanced motion detection, as well as the lower power dissipation required for battery backup operation. Aquila has developed a "breadboard" using the first generation of these processors and while initial results show promise, any meaningful evaluation is still several months away. These new processors offer significant power reductions by incorporating such functions as Ethernet and PCMCIA interfaces directly on the processor chip, thereby eliminating the power dissipation of the interconnection signal drivers of a multi-chip implementation. Furthermore, these small new processors offer a computational capacity of 9 MIPS, which is very close to that of a Pentium with significantly less power consumption and support component requirements. In addition, recent releases of CCD detectors that consume less power and are less costly are also under investigation. If these prove suitably reliable, they will both reduce power consumption and cost.

The silicon and components that provide the needed functions, computational capacity, and energy efficiency are rapidly becoming available. The algorithms and methodologies for processing the video data are emerging at the same time. The convergence of these hardware and software components will make it possible to continue to meet the increasing demands on Safeguards surveillance systems

CONCLUSION

The Safeguards requirement for verifiably uninterrupted surveillance has remained largely unchanged for a long time. However, the required implementation has changed dramatically in recent years. The need for cryptographic authentication, motion detection, advanced motion detection, interaction with other sensors and instruments, and the anticipated demand for object recognition have placed ever increasing demands on the computational performance and operational capabilities of the surveillance systems.

The new technologies developed to meet these needs have themselves often imposed additional demands on surveillance system capabilities that exceed the fundamental needs of Safeguards. For example, while Safeguards requires a motion-alarm picture to be saved only every two minutes, advanced motion detection methodologies require that pictures be processed at a rate of several pictures per second in order to ensure that the correct pictures are being saved

Furthermore, as each new technology becomes available, it fosters the development of new concepts for Safeguards applications that were never envisioned during its development. These new concepts generate new requirements that place further demands on the capabilities and functionality of the surveillance systems. These new demands, in turn, foster the further development of new Safeguards technology.

The rapid evolution of surveillance technology to meet Safeguards demands is expected to continue for the next several years as more and more of the difficult Safeguards surveillance problems are solved and resolved.