Home our services projects Face Detection on LatticeECP3-70
29 December 2012  |   Face Detection on LatticeECP3-70

Moving further with FPGA-programming domain development Rhonda Software has introduced unique implementation of Face Detection method on embedded platform. Designed solution is based on Lattice HDR-60 Video Camera Development Kit powered by ECP3-70 FPGA and process video camera stream in a pipeline mode.

Implemented Face Detector was assembled in a standalone demo capable to output processed video in 720p resolution through HDMI interface. Demo capabilities:

  • Detects up to 128 frontal faces, shows frames around up to 24 faces
  • Camera frame rate is 60 FPS, typical detection frame rate is 15-20 FPS depending on scene complexity
  • Displays number of faces detected/shown and actual frame rate (how many frames were fully processed)
  • Supports 16 detection scales (can be re-configured for more or less in order to save resources/increase frame rate)
  • HW-only implementation (does not use soft processor)
  • Does not use onboard DDR (relies on embedded memory blocks and distributed memory only)

Besides to Face Detection pipeline FPGA-demo contains full-featured character generator and includes a simple inter-frame face tracker to make improved visualization:

  • Temporal filtering – eliminates most of transient detections (typically false positives) and face frame flickering.
  • Motion smoothing – significantly reduces face frames oscillation due to tremor and slight variation of lighting / camera position.

In addition to improved visual output these two enhancements noticeably reduce the number of temporary fluctuating false-positives providing valuable addition to final results.

The areas of application for Face Detection technology are numerous, starting from smart auto-focus algorithms in photo cameras and up to face recognition based automatic access control in modern security systems. Usually Face Detection is a first stage of more specific video analytics methods as it allows focusing computationally intensive calculations on face regions only significantly reducing overall task complexity. So face detection is an enabler for audience measurement analysis that includes gender/age recognition, attention detection (through the face angle estimation), ethnicity/emotion classification and so on. As a standalone feature, Face Detection can be used in video surveillance systems where face found can play a role of trigger for high resolution recording to capture scenes of interest with more details.

In complex embedded systems FPGA-based Face Detection module will work as a highly customizable hardware accelerator relieving CPU / DSP for business-logic execution.

The latest bitstream of Face Detection demo for HDR-60 Video Camera Development Kit can be downloaded from: http://computer-vision-software.com/lattice/Face_Detection.zip


Do not hesitate to contact us for any question regarding our services. We will be glad to respond in detail.