Human Face Identification AI

Lattice sensAI Reference Design

This reference design implements Convolutional Neural Network (CNN)based human face identification on Lattice’s low power ECP5 FPGA using an image sensor. The training process is completed on a GPU-powered machine to sharpen the CNN to detect points in the human face to measure and distinguish the difference between people. This design can be used for identification of other objects by modifying the training database.

The design implemented on hardware includes an SPI, DDR memory interface blocks, an image signal processing engine, 8 CNN acceleration engines and a counting and overlay engine to show registration and identification results. It is optimized for the ECP5 85K LUT FPGA.

When the design is deployed on the FPGA, a person can register their face during the registration phase by extracting and storing 256 16-bit values representing distinguishing characteristics. In the identification phase if a registered person stands in front of the hardware 256 16-bit values are extracted and compared to the stored list of values for verification.


  • VGG8 like – 8x (Convolution, Batch Normalization) + 4x Pooling + 1 fully connected CNN
  • Network is trained with around 1 million images including augmentation for various conditions
  • Runs at 2 frames per second with 90 x 90 RGB Input
  • Total ECP5 power consumption of 850mW

Jump to

Block Diagram


Quick Reference
Technical Resources
ECP5 Face Identification Quick Start Guide
FPGA-AN-02010 1.0 11/1/2019 PDF 1.3 MB
Human Face Identification Using CNN Accelerator IP - Project Files
1.0 9/9/2019 ZIP 147.3 MB
Human Face Identification Using CNN Accelerator IP - Documentation
FPGA-RD-02062 1.0 9/9/2019 PDF 4.8 MB

*By clicking on the "Notify Me of Changes" button, you agree to receive notifications on changes to the document(s) you selected.

Like most websites, we use cookies and similar technologies to enhance your user experience. We also allow third parties to place cookies on our website. By continuing to use this website you consent to the use of cookies as described in our Cookie Policy.