20face Assignments

20face Assignments

Open assignments

Research and implementation of deploying 2x2D Cameras IP based cameras, as an alternative to 3D in facial recognition systems, in practice.

This internship focuses on investigating the practical application of two 2D IP cameras as a viable alternative to a single 3D IP (depth) camera for performing anti-spoofing. Currently we’re using FRAMOS D435e 3D IP camera’s which generate a very high (~500Mb/s) uncompressed datastream which isn’t scalable for large deployments.

We already developed an algorithm to use 2x2D USB cameras to perform anti-spoofing.. The core of the assignment is investigating and solving problems when applying this method using IP based cameras.  Things like timing issues, frame synchronization and associated drifting when capturing frames simultaneously with both 2D cameras over a network.


Additional goals could be, researching solutions for various challenges including:

  • Improving the 2x2D anti-spoofing algorithm by enhancing the detection of counterfeit faces, for instance, by presenting two photos of a face from different angles, utilizing techniques like stereo imaging.
  • Improving the 2x2D algorithm developing methods to precisely distinguish & correlating multiple faces from both camera sources. Thereby preventing confusion in face rotation, employing techniques like stereo imaging, for example.
  • Improving the performance of stereo image generation.
  • Enhancing the speed of synchronizing and maintaining image consistency.


Does this appeal to you?
Contact us at [email protected].

Research and Implementation of embedded facial recognition systems leveraging hardware accelerators like TensorFlow Lite

This internship is focused on supporting our existing facial recognition software on (ARM based) embedded systems. Eliminating the need for more expensive x86 based hardware like Intel NUC. The primary objective is to explore and implement a hardware solution that allows our facial recognition model to run on-chip, optimizing the system’s size, efficiency, and performance.


This can include:

Hardware exploration:

  • Investigate and identify suitable embedded chips capable of running our facial recognition model effectively.
  • Evaluate the compatibility of TensorFlow Lite with the selected embedded chips.

Face Detection model optimisation:

  • Adapt the existing face detection model to run efficiently on-cip
  • Explore more performant compatible alternative face detection model

Face recognition model optimization:

  • Adapt the existing TensorFlow facial recognition model to for example a TensorFlow Lite for efficient on-chip execution.
  • Explore techniques to reduce the model size while maintaining accuracy & compatibility.

Real-time processing:

  • Implement real-time processing capabilities on the selected embedded hardware.
  • Optimize the software to handle facial recognition tasks seamlessly within the constraints of the embedded system.

Performance metrics:

  • Define and measure key performance metrics such as processing speed, accuracy, and resource utilization.
  • Fine-tune the system to achieve optimal performance on the chosen embedded platform.


Does this appeal to you?
Contact us at [email protected].