CSEM unveils energy-efficient emotion recognition system MOODY

CSEM introduces MOODY, an innovative ultra-low-power vision system capable of recognizing human emotions using the energy of a single coin cell battery.

CSEM has announced the launch of MOODY, a groundbreaking vision system that can detect human emotions with minimal power consumption. This innovative system operates on a sub-milliwatt (sub-mW) power level, enabled by a single coin cell battery, making it a game-changer for human-machine interaction (HMI), access control, and driver safety applications.

Envision a world where devices can understand and react to human emotions, all while consuming negligible amounts of energy. CSEM is making this a reality with MOODY, a sensor that challenges the traditional energy demands of machine learning (ML) applications.

“Central to MOODY is CSEM’s ERGO640 high dynamic range (HDR) image sensor and Visage ML accelerator—technological marvels that empower MOODY to process data directly at the source with unparalleled energy efficiency,” explained Dr. Petar Jokic, Senior R&D Engineer at CSEM. “These components, in harmony with advanced embedded learning algorithms and resource-efficient software, render MOODY an ideal candidate for spearheading the future of next-generation ULP ML and HMI applications.”

The three pillars of MOODY

The ERGO640 high dynamic range (HDR) imager is a key component of MOODY. It uses a patented time-domain pixel-level analog-to-digital (A/D) conversion architecture, which captures a wide range of light intensities to ensure accurate imaging even in challenging lighting conditions. Consuming less than 1 milliwatt (mW) for VGA resolution at 10 frames per second (FPS), the ERGO640 is crucial for battery-operated devices. With a pixel pitch of 6.3 micrometers (µm) and a fill factor of 83%, the ERGO640’s on-chip frame memory and adaptable interfaces ensure compatibility with resource-limited microcontrollers (µC), making it ideal for energy-efficient applications.

The Visage ML system-on-chip is another cornerstone of MOODY. This intelligent processor operates at sub-mW power levels, combining a binary decision tree (BDT) engine for object detection with a multi-precision convolutional neural network (CNN) engine for complex classification tasks. This versatility allows the chip to efficiently handle various ML workloads. Supporting comprehensive end-to-end edge processing, this chip enables always-on image analysis within a sub-mW power budget, which is perfect for continuous operation in wearables and access control devices.

MOODY’s success is not just hardware-based as it also relies on meticulously optimized software algorithms. CSEM has refined two neural network (NN) algorithmic blocks: one for efficient face detection and another for emotion recognition, requiring less than 300 kilobytes (KB) and 200 KB of memory, respectively.

Innovations beyond MOODY

CSEM’s expertise in ultra-low-power vision systems and edge artificial intelligence goes well beyond the MOODY project. The company also offers eye-gaze and gesture tracking technologies to enhance safety in autonomous aviation and vehicles. Their facial recognition technology provides secure identification for access control and efficient human-machine interactions. Additionally, CSEM’s advanced vision technology with edge processing has the potential to significantly improve areas such as patient care, driver safety, and building management.