|
The use of real time video event detection or video
analytics (VA) in surveillance and related fields is increasing steadily.
The typical VA solution implements event detection using software
algorithms that run on CPUs or special accelerator boards within digital
video servers or embedded in intelligent cameras or other edge devices
(e.g. video encoders). As the demand for more complex solutions increases,
both in terms of image resolution and the speed, number and complexity of
events to be detected, so does the demand for higher performance VA
algorithms. Traditional software-only approaches can accommodate these
demands as long as they have sufficient processing power to run the
algorithms. Serious limits arise as the computational task scales up
faster than what general-purpose CPUs or Digital Signal Processors (DSPs)
can provide even when VA algorithms run on dedicated processors. Current
approaches do not scale! This results in limited capabilities, increased
cost of cameras, and excessive development effort and cost. |
|
Eutecus addresses this critical issue and limiting factor
with the Multi-core Video Analytics Engine (MVE™), an easily configurable,
flexible, compact, high-performance processing architecture enabling a new
scalable level of complete video analytics solutions within a single
low-cost FPGA chip, thus allowing new capability embedded in intelligent
surveillance installations. MVE™ combines an inherently parallel
multi-core processing architecture with unique embedded complex video
analytics algorithms in an easily configurable building block, reducing
effort for camera design and application programming, resulting in reduced
development cost and time to market. Embedded solutions also benefit from
low-cost and low-power implementation when embedding the entire MVE™ into
a single FPGA. The initial commercial implementation of MVE™ is on the
Xilinx Spartan-3A DSP 3400A™ FPGA chip.
MVE™ provides video analytic capability with “all the
rules, all the time” analytic processing at full HD 720p resolution and at
full 30 fps, enabling greater accuracy in results.
MVE™ incorporates Eutecus’ Cellular Multi-core Video Analytics (C-MVA™) processor –
composed of specialized image processing IP cores – which was developed
based on extensive research into cellular architectures that mimic the
processing found in human vision. Eutecus works closely with security
camera OEMs to incorporate MVE™ into their product lines, either as a
single-chip solution or as a Video Analytics module in high-end cameras
where a separate DSP can take care of tasks such as encoding and
compression.
In order to jump start the design of MVE™-enabled
cameras, Eutecus provides a fully functional reference design, the Bi-i™
V401X Video Analytics module, with easy-to-use PC software to set up and
configure MVE™ to perform video event detection. All this is available in
the Eutecus MVE™ Evaluation Kit V3.0, which also contains MVE™ firmware
for the XtremeDSP™ Video Starter Kit from Xilinx. Developers using MVE™
can also benefit from the Targeted Design Platform tools and environment
from Xilinx. |
|
MVE™ System-on-Chip |
|
|
MVE™ Features: |
- Stand-alone, flexible, scalable, configurable Video Analytics
engine based on high-performance
- C-MVA™
processor architecture
- Fully configurable VA functions such as people counting,
detection of abandoned/stolen objects, entering forbidden zone,
loitering, movement in wrong direction
- HD image resolution at 30 frames per second
- Reduced time-to-market due to easy system integration
- Enables system-on-chip (SoC) VA solution in Xilinx FPGAs
- Significantly reduced system cost and power consumption with
superior performance Arbitrary number of events detected in
parallel for greater flexibility
- Pixel processing implemented by parallel IP cores
- Object detection and decision making implemented using
powerful parallel processing algorithms
- MVE™ is inherently scalable for various bandwidths, addressing
a wide range of needs for now and the future
| |
| MVE™ can be easily integrated into OEM security camera design.
OEM development only needs to be concerned with overall camera
design and component integration. |
| MVE™ Integration Support: |
- MVE™ integration via standard FPGA tool chain
- Software based configuration and programmability (enough to
handle MVE™ at software level)
- Reference designs with tools and libraries available to speed
up development
- Customization and support available for integrating MVE™ into
new camera platform
|
| Read more about MVE™ in Issue 66 (Fourth Quarter 2008, pp28-32) of Xcell journal. | |
OEM camera and FPGA design fit into existing tool chains
with minimal software development effort. High-end cameras may run
additional video processing algorithms on DSP. |
