RE: SEE Tweet / GM-Tesla3 Jun 2019 17:55
Schlemiel. The video posted on twitter kicks of at about 8.25m and the driver tries to fool the system by glancing away, with Shades to test the system.
SEE post says 'This video highlights the sophisticated driver assist capabilities of @GM's @Cadillac #SuperCruise system, including @seeingmachines' advanced FOVIO driver monitoring tech which forms an integral part of the vehicle’s #safety architecture'.
They are showing us the systems sophistication.
This Patent disects why it's important to be able to distinguish between a glance event and a Fatigue event which is what they are refering to.
https://worldwide.espacenet.com/publicationDetails/description?CC=US&NR=2019122044A1&KC=A1&FT=D&ND=3&date=20190425&DB=&locale=en_EP#
[0005] Due to the position of the imaging camera relative to the driver's face, current systems encounter a problem in that distinction between an eye closure and a glance event such as a look down is difficult. This is illustrated in FIG. 1, which illustrates, in the top panel, an alert glance down at a mobile phone and, in the bottom panel, a partial eye closure due to drowsiness. From the perspective of the camera, the eyes of the driver appear to close in both situations, despite the fact that, in a look down event, the driver's eyes may actually be far from closed.
[0006] By way of example, U.S. Pat. No. 5,867,587 describes a fatigue measurement using blink detection and U.S. Pat. No. 5,878,156 describes a technique for fatigue measurement based on detecting the state of the eyes. Both methods are fragile if applied to tasks that involve wide ranging head motions, such as when driving a car.
[0007] Incorrect distinctions between look down events and eye closures can cause the driver monitoring system to incorrectly characterize the driver as fatigued or drowsy, which can lead to false alerts produced by the operator monitoring system. The incorrect distinctions can also skew long term characterizations of that driver that are derived from historical monitoring data.