Japanese driver monitoring specialist PUX Corp has released a new version of its software designed for driver monitoring during the Covid-19 pandemic. The new version is able to monitor drivers effectively even when they are wearing face masks as required by Covid-19 regulations. The new version will be licensed to automotive device suppliers and automotive supplies vendors from December.

PUX has been developing the image recognition software for driver monitoring since 2013. Its “FaceU for DMS” has been used in passenger cars both in Japan and overseas since 2018. Around 430,000 cars, comprising nine vehicle types made by four manufacturers carry the technology.

The software detects driver status such as “Looking aside“, “Distraction“, “Drowsiness“ and “Posture collapse“. Driver monitoring software (DMS) became common due to the EU NCAP programme and the trend towards assisted and autonomous driving.

Recently, the advent of COVID-19 has led to many drivers having to wear face masks. This has affected the accuracy of facial detection, and has also impeded the identification of drivers.

Facial detection is able to identify the state of the driver, whether they are drowsy, ill, distracted or otherwise impaired. Facial identification establishes the identity of the driver as a specific individual.

To address these issues, PUX developed FaceU for DMS Ver.2.8.2, which can carry out facial driver detection and identification even when drivers are wearing a variety of face masks.

PUX achieved facial identification and detection despite a driver wearing a face mask by combining statistical analysis methods with a newly developed deep learning rapid inference engine.

PUX says that its new inference engine does not require special GPU, FPGA or NPU hardware. Instead, it says its engine can run and execute real time processing on relatively inexpensive CPU, such as ARM Coretex-A7 1GHz x 1 core.

Recognition can be carried out across a wide range, with a yaw of 180 degrees and a pitch of 110 degrees. The company says its engine can recognise low quality images in brightness, blur and high levels of image noise with a high degree of accuracy.