Face recognition package increases accuracy
Image sensing libraries used in the OKAO Vision face recognition package by Omron Electronic Components Europe are claimed to provide “highly accurate” deep learning face recognition. Developers can deploy OKAO Vision on their choice of embedded hardware platform.
The deep learning libraries of OKAO Vision Face Recognition V9.0 address applications that require accuracy under various conditions including poor lighting and when the face is at various angles relative to the detector. These include security and access control, time and attendance monitoring, login/wake up systems, and camera auto focus/auto-exposure control.
The platform can be used to monitor attendance at face to face and online meetings, which will facilitate contact tracing and verification of attendance. Another application will be in automotive design, for example in driver recognition to manage features such as seat adjustment.
The face recognition libraries achieve “excellent” evaluation results with various skin tones and face sizes, says Omron. It delivers a low error rate down to image size as small as 40 pixels. Benchmark testing with Intel and Arm processors has demonstrated that OKAO Vision Face Recognition V9.0 maintains exceptionally fast recognition times despite the enhanced accuracy, reports Omron. This ensures that users in access control applications for example will be barely conscious of the need to wait for validation of their identity.
The complete OKAO Vision Face Recognition V9.0 package contains modular libraries that provide sensing capabilities including expression estimation, age and gender estimation, and photographic image beautification including red eye reduction, facial shaping, eye enlargement, and blemish removal. Users can combine various modules’ functionalities to add value to applications.
OKAO Vision is available as a set of software libraries and can integrate with Linux, Windows and iOS operating systems. Users can leverage Omron’s machine vision package in embedded systems running on custom hardware. Off-the-shelf libraries are already available for various platforms, adds Omron.