6
and decreasing the price are very important factors for
both developers and system
producers. Most of the ADAS solutions are camera-based and are using video processing
for monitoring either the road or the driver to detect abnormal behaviors during driving.
By fusing various sensors (cameras, radars, laser scanners, etc.) the field of view is
enlarged, and the perception precision of objects in the relevant
regions for the driving
process is significantly increased [11]. Fusing data and information with different levels of
quality and sampling rates is another challenging issue in ADAS technology. Darms
et al.
have proposed a modular system architecture for sensor data processing and combining
short range sensor, long range sensor, video information, actuators feedbacks, and vehicle
dynamics sensors in ADAS technology [12].
In general, ADAS has three data processing
levels which are sensor level, fusion level and application level. Figure 1
does show the
abstract architecture of an ADAS system with respect to the different processing levels.
Figure 1-1: ADAS processing levels architecture
The following list does give a sample of functionalities provided by different ADAS
solutions involving visual information and a digital camera:
Lane departure warning system (LDWS)
Traffic sign recognition
Pedestrian detection
Fatigue detection
Adaptive cruise control
The large amount of data provided by cameras requires a huge processing effort.
Speeding up the processing is therefore extremely important especially while facing real-
time constraints [13]. Some algorithms such as stereo vision
and depth estimation are
particularly demanding in terms processing. Some algorithms have dependency and it is
