Before we leap headfirst into this announcement with gusto and abandon, we first need to understand some of the underlying concepts. The solution discussed here revolves around something called Video Content Analytics. In the context of this announcement, this refers to the intelligent analysis of video content by a computer to provide real-time information about some event or events that are occurring in the real world.
Video Content Analytics differs from simple Motion Detection, which generates an alert when motion is detected (perhaps some defined number of pixels changing attributes in some defined area of the image), but which has no concept of object attributes like “car versus person versus animal”, trajectory, entering a forbidden zone, and so forth.
Video Content Analytics also differs from traditional Pattern Recognition Technology, such as license plate look-up or facial recognition, which is typically performed as a post-processing function sometime after the video has been recorded.
The simplest way to look at this is that Video Content Analytics is based on a suite of algorithms running fast enough to be able to detect movement in the image, decide whether this movement is associated with something of interest (is it a car or a person or an animal and is it doing something it shouldn’t be doing), and – based upon user-defined rules – determine if this event (or combination of events) is of sufficient interest to alert the operator(s).
The first point to note is that Altera is announcing the world's first FPGA-based full-HD 1080p/(30 frames per second) 30fps video analytics on a Cyclone IV FPGA. As opposed to working only with standard video (4:3 aspect ratio, 704x480 resolution, low bandwidth), this solution works with full HD video (16:9 aspect ratio, 1920x1080 resolution, high-bandwidth).
Altera says that their single-chip solution provides a new level of video analytics performance, combining exceptionally high throughput (60-Mpixel per second) with pixel precision detail not possible with traditional digital signal processing (DSP)-based approaches. The solution includes Eutecus' Multi-Core Video Analytics Engine (MVE) intellectual property (IP), which performs the analytics functions in the FPGA.
Because this solution can be easily integrated into HD Internet protocol cameras, it is ideal for a variety of applications, including traffic surveillance that monitors accident detection, vehicle counting, lane-exit detection, stopped traffic, red-light violations and vehicles moving in the wrong direction.
Consider the image below, which provides a simple example of the detection of turning movement using a composite event. First of all the operator has defined two areas of interest shown as E1 (“Event 1”) and E2 (“Event 2”) in this image. The idea is that if a car first enters area E1 and subsequently passes through area E2, then the system can identify the fact that the car has turned left.
When I first saw this I wasn’t quite sure how it worked. It seemed to me that one car could enter area E1 thereby triggering the first event, but then another car (possibly of the same color) could appear in the scene moving from right to left passing through area E2, thereby confusing everything. It turns out that the Altera-Eutecus system is far more sophisticated than this; once a car has been detected in area E1, that specific car will be tracked as it leaves E1 and moves throughout the scene.
Now, although this may not seem tremendously exciting at first, it becomes more so when you discover that you could have multiple cars in the “pipeline” having passed through one area with some of them heading to the second area and others going somewhere else, possibly passing through other defined areas. Speaking of which, the operator can define an unlimited number of areas (and associated events) in a scene with an unlimited number of relationships between the various areas.
And things become even more sophisticated, because the system can be set up to differentiate between cars on the road and people on the side of the road (or people on the road where they shouldn’t be) and so forth. Also, the system can account for the fact that things change size, getting bigger when they are in the foreground and smaller as they move away from the camera. Furthermore, the system detect and ignore things like rain and fog and suchlike which would make simply motion detection systems “throw a wobbly” as it were.
Altera says that this FPGA-based video analytics solution is the first and only solution capable of delivering the processing power required for full HD 1080p/30fps real-time video processing. Also that the tremendous performance provided by the massively parallel FPGA implementation – coupled with the use of multiple soft core microprocessors (as discussed below) – provides the ability to track dozens of user-defined rules. For example, the user can define rules and configure alerts for specific events they wish to be detected, such as a person entering a room or a restricted area in an airport terminal or a building entrance. (All of this can be remotely defined and updated via the Internet).
This form of Video Content Analytics offers surveillance customers the ability to monitor videos automatically. Consider the security room associated with a large building – the sort of thing you see on films with folks constantly scanning tens or hundreds of video screens looking for something untoward happening. Now consider the same system augmented by Altera’s Video Content Analytics system – in addition to alerting operators of “events of interest” in real-time, this system may also eliminate the need for people to scan hours of footage “after-the-fact” to detect and determine specified events.
The combination of Altera's Cyclone IV FPGAs and Eutecus' MVE IP offers video analytics designers greater flexibility to customize their applications. The IP combines massively parallel algorithms and specialized coprocessors with multiple Altera Nios II soft microprocessor cores integrated into the Cyclone IV FPGA. The IP also comes with a software GUI that allows designers to customize the event-detection parameters and rules for their specific applications. Eutecus' IP includes a robust application programming interface (API) that allows customers to interface with their own video management systems or develop their own custom GUI.
The performance and flexibility of Altera's video analytics solution offers customers a unique benefit by eliminating the need to make tradeoffs between system performance and the number of rules running at the same time.
Altera also offers a single-vendor source model for the video analytics solution. The IP and FPGA can be purchased directly from Altera, which simplifies the development of a video surveillance system significantly. The purchasing model eliminates the need to work with multiple vendors for licensing fees, NRE charges, or royalties.
"With Altera's FPGA's unprecedented video processing capabilities coupled with Eutecus' MVE analytics technology, customers can track multiple objects with multiple rules in varying weather conditions in real time, giving customers a more powerful and flexible solution compared to existing technology. This solution can also be leveraged for other applications such as automotive driver assistance which includes rear view and surround view cameras," said Arun Iyengar, vice president of the military, industrial, and computer, consumer, and storage business units division at Altera. "Altera's one-stop shop approach for the FPGA and IP reduces the barriers for customers looking to expand their analytics capabilities."
"Altera's Cyclone IV FPGA and its Quartus II software offer the ideal design environment for Eutecus' MVE solution," said Csaba Rekeczky, CEO and president of Eutecus. "Video analytics customers can take advantage of the partnership between our companies by leveraging Eutecus' highly modular, flexible and scalable IP product and Altera's high performing and customizable FPGAs."