Face Centric Analytics
February 26, 2020

We Teach machines the beauty and complexity of human faces

80% of all pictures taken today contain at least one face.

Simply put, face analytics are the backbone of computer vision. Technologies surrounding exposure, focus and white balance calibration, biometrics, portrait enhancement or emotion recognition are the bridge from today’s computer vision-centric cameras to tomorrow’s decision making intelligent cameras.

DTS offers a hybrid and scalable software/hardware implementation in a flexible architecture to enable intelligent imaging capabilities in any embedded device. This enables the best mix of low-power, high performance and state-of-the-art functionality that delivers outstanding user experiences.

Driven by convolutional neural networks (CNN), our face-centric analytics pack contains FDX (next-generation face detection technology), Face Features Detection (FFD)Face Classification (FC) and Emotion Detection.


  • Face Detection & Tracking solution based on a CNN detection engine.
  • Neural network enhanced parametric model for human faces.
  • Demographic-friendly CNN empowered face classification technology.
  • Neural based emotion classification with easy adoption of specific classes.

FDX is a novel implementation of a face and facial landmark detector entirely designed to run optimally on mobile platforms.

Based on a CNN architecture specifically designed to cover all corner-cases, the system is able to detect faces in an unconstrained environment.

FDX employs CNNs designed for multi-task face analytics able to simultaneously detect faces, perform regression for improved face framing, detect facial features (eyes and corners of the mouth) and output the orientation angles.


BALANCED and calibrated speed, quality and locking time for UX and power requirements improvement.

FAST re-detection in video mode on new frames for increased stability in video sequences. Temporal filtering of outputs is performed for increased consistency.

FLEXIBLE and configurable face locking, single or multi-frame to optimize between power consumption and speed of detection.

EASY control over detection and false positive rates to balance solution quality for use case requirements.

FDX has a flexible and efficient code architecture for CNN inference designed with parallelization in mind.
CNN execution using NEON optimization and configurable number of threads.
CPU only execution. No GPU required.

DTS’s face features detection solution uses two deployment modes with 18 or 91 points models for fast or quality-centric real-time processing.

The solution also provides three axis face orientation estimation with yaw support [-90° : +90°]. It uses FDX landmarks and improves accuracy.

DTS’s face classification and emotion detection solutions offer classification for gender, age and expression in tracking mode. Seven emotions detection work in tracking mode as long as both eye positions can be tracked.

There is easy adaptation to specific classes per use case need (e.g. FACS) while remaining core engine agnostic to camera system (supports visible, NIR).