MulticoreWare joins forces with Melexis for an AI algorithm development. They developed face understanding algorithm modules such as face detection, face recognition, drowsiness/distraction detection and anti-spoofing detection by use of a Melexis EVK75027 ToF sensor.
MulticoreWare enhanced the internal data annotation tool by utilizing distance images to label precise face key points. Using this information, a custom annotated dataset was created and used to train neural networks for face identification and recognition. The system is able to operate accurately and reliably in diverse illumination settings.
The partnership demonstrates the effectiveness of AI using ToF cameras to achieve robust performance for a wide range of in-cabin applications such as driver authentication, drowsiness detection, driver attentiveness, etc.
“It was a pleasure collaborating on this joint demonstration. It’s exciting to conclude that existing 2D neural networks perform well on ToF confidence images without model retraining. This ensures system integrators can switch technologies with minimal effort. The ToF distance images improve solutions with high reliability, resulting in truly safe and robust end-user applications,” said Kristof Lieben, Melexis Product Manager.
“Melexis is an obvious partner for us with their long-term leadership in reliable embedded sensors for the automotive sector,” said Jayesh Patel, VP & GM, Autonomous Vehicle & Automotive business Unit, MulticoreWare. “The joint demonstrator clearly illustrates that time-of-flight-based data processing leads to creative and efficient in-vehicle features.”
