Movidius weds thermal imager and computer vision

Movidius weds thermal imager and computer vision

Technology News |
By Julien Happich

“We are putting together those two separate sensing modalities into one by collaborating with Flir systems,” Movidius CEO, Remi El-Ouazzane, told EE Times. Movidius is a developer of low-power machine vision technology, while Flir is a leader in thermal imaging technology.

Flir’s Boson

The companies have worked together for two years. Their efforts resulted in adding Movidius’ advanced computer vision capabilities to FLIR’s newest thermal imaging camera core called Boson.

More specifically, Boson integrated with Movidius’ vision-processing SoC called Myriad 2 VPU can now implement “advanced image processing, super resolution and noise filtering,” according to El-Ouazzane. Compared with Flir’s previous generation thermal core called Tau 2, the new module is 10 times smaller and lighter in weight, while consuming only half the power.

“The cherry on the cake is that there is still enough processing power left on this hundreds of gigaflop chip” that’s sitting next to the thermal sensor, said Movidius CEO. That can be used for additional image processing and analytic algorithms.

In short, aside from using the new module for the creation of thermal images, Flir’s customers can use it for everything from a vision analysis suite to porting their own vision algorithms, he explained.

No one else has yet combined a thermal image sensor with a vision processor. “Nothing like this exists today,” boasted El-Ouazzane.

Pedestrian Counting: Thermal imaging is well-suited to the task as it performs well in various weather/light conditions, in addition to being able to easily differentiate the background from live subject through heat signatures. (Source: Movidius)

Movidius’ Myriad 2 plays three roles inside Boson.

First, it processes raw imaging data from the thermal imaging core, according to El-Ouazzane. As Pierre Boulanger, Flir’s CTO, pointed out, “In infrared, no two pixels are created equal. We have to process every single pixel at many different levels—way more than a visual sensor.” The highly programmable Myriad 2 “allows us to run those special algorithms,” said Boulanger.

Second, Myriad 2 in Boson performs advanced computer vision analysis.

Third, Myriad 2 acts as a system-level SoC for the complete system, noted El-Ouazzane, eliminating “three other subsystems.”

Boson users no longer need to add such subsystems as input/output, display drivers to monitor graphics and power, and visual analytics, El-Ouzzane added.


Broad applications
Thermal imagers can detect radiation in the long-infrared range of electromagnetic spectrum. This enables thermal imaging cameras to see warm objects against cool backgrounds—day or night. Its applications range from military equipment and surveillance to automotive night vision and firefighter helmets that can see through smoke.

Once a thermal imaging core is combined with a visual intelligence processor core like that of Movidus’ Myriad 2, thermal imaging cameras can not only get smarter. They can also double as embedded cameras that capture images using visible light.

Drones today are already carrying thermal image sensors for security, surveillance or inspection purposes. DJI, the world’s leading maker of drones and aerial cameras based in Shenzhen, China, offers, for instance, a drone equipped with Flir’s Tau 2 thermal imaging core.

Last month DJI launched its Phantom 4 aircraft, with Movidus’ Myriad 2 vision-processing SoC in it. The Phantom 4 is said to be capable of sensing and avoiding obstacles in real time and hover in a fixed position without the need for a GPS signal.

The Movidius CEO said, “I can’t speak of either Flir or DJI, but it is conceivable that DJI can now use Flir’s new Boson core,” employing a combination of night vision and visual intelligence processing.

Boson is a configurable thermal camera core that works in VGA and QVGA resolution. Its camera body is 21 x 21 x 11 mm, weighing 10 grams. Its low power consumption starts at 500mW. It offers built-in support for physical- and protocol-level interface standards such as Ethernet, USB, SD card, LCD interfaces, and wireless video.

Boson is a platform-based solution allowing different configurations at different prices, performance grades and image quality, according to Flir.

Computer vision algorithms allow systems to understand how individual features move from frame to frame. These basic building blocks support object detection & tracking, pose estimation and gesture recognition. More advanced neural networks are even beginning to understand very subtle cues such as moving with a concealed object, sentiment and erratic behaviour. (Source: Movidius)

About the author:

Junko Yoshida is Chief International Correspondent at EE Times

Related articles:

The future of video surveillance: HD, hyperspectral and stereoscopic

IMEC hyperspectral image sensor exceeds 140 bands

IR cameras soon a commodity, warns Yole


If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles