Optoelectronics
Total # of frames No head track % Innacurate head tracking % (> 3cm error)
2200 0.14% Table 1. Head-pose accuracy results
With a single MLX75027 sensor, which has VGA (640 x 480 pixels) resolution, the resulting DMS can track both driver and passenger. The sensor has a minimal footprint, permitting a cost-effective and compact module that can be easily implemented into the rear-view mirror assembly. Melexis has a range of automotive-qualified ToF sensors from QVGA (320 x 240) resolution to VGA, which are suited to DMS applications.
Figure 2: 3D point-cloud image constructed using ToF-sensor data
in parallel to create the depth image. In practice, a ToF sensor uses active light modulation at a frequency in the range 20-100MHz. Four waveforms are generated in quadrature to drive a VCSEL (Vertical Cavity Surface-Emitting Laser) or LED emitter. A lock-in amplifier calculates the four phase-shifted signals from each pixel and can distinguish signals with a known carrier wave against a noisy background to ensure accurate depth detection in all types of ambient lighting. The sensor subsystem also contains beam-shaping optics containing a wide-aperture lens, and a host microcontroller.
Using the depth data from each sensor pixel, and considering the lens parameters at the pixel location, a 3D point-cloud image can be created (figure 2).
Implementing a 3D ToF Solution Melexis partnered with 3D technology companies to create DMS demos based on Melexis MLX75027 3D Time-of-Flight (ToF) camera sensors.
With Eyeware’s attention-monitoring technology, the DMS provides eye-gaze and head tracking with high detection accuracy in a wide range of driving conditions including bright sunlight.
The algorithms use proprietary strategies based on machine learning, which ensures fast response times and low power consumption. By combining the strengths of the ToF sensor, with its internal signal processing, and the software, the system can monitor a very wide range of head movement and localise eyes well at various resolution settings.
3D ToF sensing in practice The prototype DMS, built using Melexis’ EVK75027 ToF-sensor evaluation board, was tested in laboratory conditions designed to replicate the distances and gaze angles in a real car. Experiments were carried out using a set of 15 participants. The image of a car interior, with visual targets appearing to be outside, were projected onto a screen ahead of the participant. To be able to collect eye gaze ground-truth information on the exact eye gaze, dots were shown on the screen at random locations and the participant asked to look at the specific dot and confirm. After each confirmation, the next gaze point was presented on the screen. Participants were recorded twice, for a duration of 120 seconds per session. To evaluate the head-pose detection performance, instances where head-tracking
was achieved with position error less than 3cm were regarded as successful. Accordingly, the positive detection rate achieved was 99.7 per cent, as shown in table 1. Statistical analysis of head-pose results showed an average position error of 7.3mm, with a standard deviation of 4.8mm. With the total failure rate under 0.3 per cent for very wide viewing angles, the results suggest that a DMS using ToF technology is capable of always keeping track of the face.
To assess eyes-off-the-road detection, ground-truth data points were selected based on the definition of eyes-on-the-road and eyes-off-the-road zones (figure 3). Performance was measured using “Precision” and “Recall” metrics. Precision is defined as the ratio of true positives (frames correctly detected as eyes-off- the-road) to the number of detections. A precision of 100 per cent means that no false alarms are recorded although some events may be missed.
Recall is the ratio of true positives (frames correctly detected as eyes-off-the- road) to the number of events/alarms. In the extreme case, a recall of 100 per cent means that all events are detected, with some false alarms.
The DMS detected 95 per cent (recall) of the eyes-off-the-road events. The number of false alarms (precision) was below 5 per cent.
Because eye closure is a strong indicator of driver drowsiness, eyes open/ close detection was also monitored. Over 95 per cent accuracy was achieved, despite the sensor’s relatively low VGA resolution. This demonstrates robust eye-closure detection, accurate enough to be a component of a broader drowsiness detection system.
Conclusion
DMS implemented with ToF sensors can increase safety and enhance engagement with autonomous-driving systems. A prototype DMS containing a ToF camera sensor with head-monitoring and gaze- tracking software has demonstrated high accuracy and reliability to meet new vehicle legislation coming into force from 2022.
Figure 3. Eyes on the road (green) and off the road (red) regions
www.cieonline.co.uk
www.melexis.com Components in Electronics September 2022 35 0.14% Total failures 0.28% Positive detection 99.7%
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62
