FRWY LS1046A TP Machine Learning/Video Analytics

Document created by Fabian Ramirez Employee on Mar 31, 2020Last modified by Fabian Ramirez Employee on Apr 2, 2020
Version 3Show Document
  • View in full screen mode



The following demo is part of a bundle of examples that shortly will be available in the LSDK.
As can be seen in the video, there is a menu for the user to select the desired demo and start testing the different approaches the FRWY LS1046A TPU is capable of executing.
These examples aim to demonstrate AI/ML capability through multiple “dockerized” demos including face recognition, Industrial safety & security -MPU and TPU versions-, People counter using the CPU and a pose detection demonstration. By using the native PCIe 3.0 of the LS1046A MPU, is possible to showcase the ML frameworks offload and acceleration via Coral Tensor Processing Unit.
The below video shows one of the ML examples for pose detection of a person by using a simple USB camera plugged into the LS1046A FRWY board. Using the ML frameworks from Coral Tensor, the body movements and positions are detected by using the joints as the reference for movement detection of the body.
As in the video is mentioned, the performance enhancement by using the TPU versus the CPU, improves significantly the frames per second from ~5 fps to ~30 fps.


Block Diagram