IEEE Consumer Electronics Magazine - March/April 2021 - 11

Figure 3. Different lighting zones illuminated by
different LED particles.

zones are illuminated by the corresponding LED
particles, as shown in Figure 3. The illumination
range of right high-beam is the same as that of
the left high-beam. If there are vehicle headlamps detected by the received visual images,
the corresponding illumination zones are turned
OFF to avoid glare interference.

Vehicle High-Beam Hardware Composition
The composition structure of vehicle highbeam hardware, as shown in Figure 4, is proposed
for adaptive road illumination. A camera is
installed on the vehicle front and used for acquiring dynamic visual images of oncoming and samedirection vehicles. The acquired visual images are
transmitted to the image processing platform
composed of graphics processing unit and central
processing unit processors. The oncoming and
same-direction vehicles are detected based on the
proposed image-processing algorithm, which can
be sent to the designed lamp controller via the
RS232 port. In the lamp controller, the visual
image processing results are mapped to the
corresponding LED particles according to the illumination zones of high beams. LED controlling
commands are sent to the left and right headlamps via the LIN bus. The entire 16 LED particles
can be independently turned ON or OFF, whose
brightness range can be adjusted from 0% to
100%. Furthermore, turning ON or OFF commands
can be sent back to the lamp controller via the LIN
bus as well.

March/April 2021

Figure 4. Composition of vehicle high-beam
hardware.

Multiple Vehicle Detection Under Visual Images
Vehicle lamp detection can be divided into
headlamp detection and tail lamp detection, as
shown in Figure 5. Assuming that the headlights
and taillights are relatively located in the
acquired visual images, the headlights are bright
and approximately symmetrical, whereas the
taillights are bright and have a large red component. The highlight area of gray-scale threshold
segmentation image is determined by the maximum interclass variance method. Combined
with the isolated noise cancellation, the region
linked lists of lamp lights, background spots,
and street lights can be obtained. Integrated
with vehicle lamp morphological analysis, the
vehicle headlights are detected under the constraints of connected area, central coordinate,
horizontal distance, and reference distance.
Vehicle taillights are processed in the color perception space. Based on the image gradient,
angle, and gradient histogram parameters, the
image feature pyramid is obtained. Using the feature model obtained by boost training, the image
feature pyramid is performed by scanning the
sliding window. The maximum suppression
greedy algorithm is used to merge vehicle taillights. If vehicle headlights and taillights cannot
be matched for three consecutive frames, they
are judged to be in the image edge or missing.

IEEE Consumer Electronics Magazine - March/April 2021

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - March/April 2021