Big Data-in-Motion Solution
Written by Alex Philp
GIF 2012 Volume: 10 Issue: 1 (February)
Our most important mission as members of the GEOINT community is to extract meaningful geographical information from data streaming in from sensors so we can deliver actionable intelligence to warfighters where and when they need it. Increasingly, “when” means “now.” To make this possible, we must process and analyze the data in real time—while it’s still moving.
Because the data is streaming in like a tsunami that threatens to inundate us, this “big data-in-motion” issue is becoming one of the most critical challenges we face. We are drowning in data, and every new imaging satellite, aircraft and UAV only adds to the GEOINT deluge.
We celebrate breakthroughs in spatial resolution and hyperspectral content, and we cheer faster communications links with the sensors. But while valuable, these enhanced capabilities also make the data sets more challenging to transmit, manage, archive, process and analyze. Gigabytes of data once seemed large, but now terabytes are common. We are dealing with peta-, exa-, and zetta-scale data problems, and their speed of arrival from sensors is now measured in minutes and seconds instead of days and weeks.
As if high-velocity, high-volume data weren’t significant enough problems on their own, the variety of data is increasing as well. The sub-meter imagery and full motion video that we traditionally associate with GEOINT are being fused with ELINT, SIGINT and MASINT data from many types of ground-based mobile and fixed-location sensors. Information-rich raster imagery feeds are now being cross-referenced with acoustic signals, biometric signatures, building control updates and cellular traffic—none of which are the structured data that easily fits into relational databases for traditional querying.
Big data-in-motion, therefore, is a problem that is uniquely complicated because the incoming data is high-volume, high-velocity and high-variety—increasingly referred to as “3V” data.
Defense/intelligence is just one arena facing a tsunami of 3V data and pursuing a big data-in-motion solution. In the private sector, for example, industries such as energy, utilities and telecommunications see very similar challenges as they protect their respective critical infrastructures.
It is important to note that sensors protect more than just physical assets. Cyber-infrastructure is also being monitored by sensor networks that add more data to the mix. If the past 10 years have been about identifying, mapping and assessing our nation’s critical infrastructure and associated vulnerabilities, the next decade will be spent dealing with chronic, persistent cyber-attacks on those facilities. This vulnerability is poorly understood and consistent with the 3V dimensions.
Fortunately, progress is being made as a result of so many professions and industries dealing with the same challenge. Real-time analytical processing (RTAP) of big data-in-motion exists today, but there is plenty of room for advancement in the technology. By necessity, however, RTAP development will never be “finished.” It must constantly evolve to keep pace with 3V data, which shows no sign of slowing down.
RTAP Today
RTAP technology applies computationally intensive algorithms, which perform traditional GEOINT processes, such as feature detection, pattern recognition and change detection to data sets. But rather than wait for the sensor data to be transmitted from their remote location and stored in a static database, the algorithms conduct hundreds of thousands of calculations in fractions of a second as the data streams in from the sensors.
By removing the database from the equation, RTAP technology has focused on finding new ways for analytical processing to be carried out at vastly accelerated rates in the compute memory of the chip. There are currently several approaches to this type of solid-state processing, but most use a system-of-systems method that involves a hybrid collection of hardware, firmware and software. This hybrid approach to computer architecture typically relies heavily on parallel processing to perform the extensive calculations in the CPU rather than in the database.
By not waiting for this data to stream into and come to a stop in the database, RTAP makes it possible for analyses to be faster than ever before, enabling information to reach decision-makers in minutes or seconds. Just as importantly, it eliminates the vast amounts of power and bandwidth that would otherwise be consumed in the transmission and storage of raw data.
A good example of how RTAP is used now can be found in the surveillance arena. Acoustic sensors have been buried along the perimeters of sensitive facilities to detect the approach of potential threats by continuously collecting sounds from the environment. A processing engine located nearby instantly analyzes the acoustic signals as they stream in from hundreds of sensors in the network. Embedded algorithms categorize the noises as mechanical, biological or anomalous to determine if they warrant further observation.
If a noise commonly associated with a possible threat, such as a vehicle motor, is detected in a location where it shouldn’t be, the processing system performs several functions simultaneously. It pinpoints the location coordinates of the noise on the sensor network and sends an alert in the form of an email or alarm to designated personnel who can formulate an appropriate response.
Simultaneously, the processing engine uses the primary sensor information to trigger activation of a secondary sensor, such as a surveillance camera, to train on the sound location and provide real-time video to the security command center. This provides verification and validation. As a result, we have detection, classification, localization, tracking, correlation, verification, validation and communication all occurring in network real time.
The key to the instantaneous aspect of this application is that RTAP ignores the unimportant torrent of background noise and separates out the critical pieces of data. The processing engine focuses upon the anomalous sounds, identifies them to some level, and delivers actionable information in the form of an email text alert or video feed directly to the human decision-maker in a matter of seconds. No other resources—human or automated—are wasted sifting through the terabytes of mundane acoustic signals from the sensor network.
Analysis on the Platform
RTAP research is focused on improving several aspects of the technology. Specifically, the goal is to accelerate and expand the ability to perform algorithmic calculations within the compute memory. One of the techniques being developed to accomplish this involves moving the processing and analysis physically closer to, or embedded within, the sensors themselves.
An example of how the geospatial industry is heading in this direction comes in the latest generation of digital imaging sensors that fly aboard observation satellites, aircraft and UAVs. Twenty years ago, raw data was transmitted from the satellite or delivered on a hard drive from the aerial platform to a ground facility for processing and analysis. Today, much of the pre-processing occurs on the satellite or aircraft, so that imagery is delivered to the ground station for enhancement, interpretation, fusion, change detection and a dozen other analyses.
With RTAP, we want to perform all of the processing and analysis on the platform, and possibly within the electro-optical fabric of the sensor. What would this mean? Imagine a classic GEOINT scenario involving an imaging sensor aboard a space or airborne platform. As the sensor is collecting image data, multiple algorithms are instantaneously sorting through the data searching for a pattern, feature or change in ground conditions that matches a predefined mission objective. And as with the real-life acoustic example above, the future RTAP may involve multi-sensor communication in which one type of sensor detects a feature of interest and a second sensor identifies it.
Once the target has been detected and possibly identified, the RTAP attaches three-dimensional coordinates to it and sends a communication in the form of an image chip, email or other alert directly to the warfighter positioned to act upon it. This happens within seconds of the initial target observation by the primary sensor. The communication carries only the information needed by the warfighter to make an informed decision.
The concept of embedding processing and analysis engines into GEOINT, MASINT and SIGINT sensors will be possible only if major advancements continue to be made in the hybrid computer technologies referenced earlier. New architectures in hardware, firmware and software are part of the equation for RTAP success, and high performance computing and cloud databases will play important roles. But the most important need right now is a fundamental shift in the way algorithms are developed.
As computer technology evolves to become faster and more scalable to push the boundaries of computationally intensive algorithms, processing will continue to move away from serial to parallel architecture. Parallel processing appears to be the only solution to scale beyond the limits of existing systems. This means that algorithms must be written for parallel execution—a sea-level change for most code developers in GEOINT and other big data industries.
The advancements described here are by no means insurmountable. Based on the existing rate of progress and new technologies coming online, RTAP has reached an inflection point that may soon put it in front of the big data-in-motion tsunami, revolutionizing the delivery of actionable intelligence to warfighters in support of the GEOINT mission. ♦
Dr. Alex Philp is the founder and chief executive officer of TerraEchos, which develops solutions for big data-inmotion challenges for monitoring and security applications. This e-mail address is being protected from spambots. You need JavaScript enabled to view it
