3D
RECONSTRUCTION
AT THE EDGE WITH
AI
DESIGN YOUR OWN PROGRAMMABLE STEREO CAMERA AT THE EDGE
WITH A BOM COST
LOWER THAN YOU THINK
CPU
Intel® x86/x64 Linux
or NXP i.MX 8M Plus
or Raspberry Pi CM4
AI ACCELERATOR
Hailo-8™ from Hailo
3D RECONSTRUCTION
software library from Deep Vision Consulting
CAMERAS
your preferred camera modules
with hw-sync features
STEREO
choose baseline and lenses according to your use case
3D VISION
create your own logics or get custom developments from Deep Vision Consulting
SOFTWARE LIBRARY
FOR 3D RECONSTRUCTION
AT THE EDGE
3D metric point clouds can be created from stereo cameras by correlating the differences between left and right video streams: this task is commonly known as stereo matching. State of the art Deep Learning techniques for stereo matching are computationally very demanding and the neural architectures are quite complex too: for all these reasons, running them on Hailo-8™ hosted on an embedded platform is not just a piece of cake.
Leveraging the fl exibility of Hailo’s neural core, Deep Vision Consulting created a software library for 3D reconstruction at the edge with stereo vision, by customly designing a stereo matching Deep Learning model containing 3D operations, like 3D convolutions, and porting it to Hailo-8™ hosted on NXP i.MX 8M Plus or Raspberry Pi CM4. See performance in table.
Unleash the potential of 3D Vision at the Edge using Hailo-8™
3D Vision + Deep Learning + at the Edge
Computer Vision is not just about 2D Vision: cool but just... flat! On the other hand, 3D Vision means having depth, point clouds and metric data, that are necessary in many use cases from industries like Automotive, Robotics, Agritech, Industrial Automation and others.
Approaching 3D vision with Deep Learning techniques is possible and unleashes great opportunities; however, technical and computational complexity increases signifi cantly. And when Deep Learning for 3D Vision needs to be applied at the Edge, the matter can get extremely intricate.
Hailo and Deep Vision Consulting
for 3D Vision at the Edge
The potential of Hailo-8™ applied to regular Computer Vision tasks is clear, but what about 3D Vision? Tera-ops are defi nitely not missing, but it’s not enough. To hit the road with Hailo-8™ for 3D Vision at the Edge, it’s necessary to have familiarity with architectural design of Deep Learning on one side, and in-depth understanding of Hailo-8™ acceleration features and its toolchain capabilities on the other.
After having solved 3D Vision tasks via Hailo-8, taking it out of the standard operation specifications, Deep Vision Consulting is now Hailo software partner for 3D Vision.
With more than 10 years of experience in producing 2D and 3D Computer Vision solutions, Deep Vision Consulting has a long list of success stories brought to production-grade implementation on embedded systems such as NXP iMX6/8, Xilinx Zynq/Ultrascale+ and Raspberry Pi, just to name a few.