Research on the self-driving and on the autonomous vehicles field is constantly increasing.
The administrations are putting all their efforts to raise awareness among
drivers about the responsibility that driving a vehicle implies. By their side, researchers
are working hard to make vehicles as safe as possible in order to decrease
the number of fatal accidents in our roads. Around 30% of these accidents have a
key factor: speeding. Since some decades ago until our days, speeding has become ...
Research on the self-driving and on the autonomous vehicles field is constantly increasing.
The administrations are putting all their efforts to raise awareness among
drivers about the responsibility that driving a vehicle implies. By their side, researchers
are working hard to make vehicles as safe as possible in order to decrease
the number of fatal accidents in our roads. Around 30% of these accidents have a
key factor: speeding. Since some decades ago until our days, speeding has become a
widely investigated topic by the large automotive companies. The European Union
(EU) has recently announced new measures that will be taken in 2022 aiming to
reduce car accidents. The most relevant one, is the measure that forces new vehicles
manufactured at that year to have an intelligent speed assistance device integrated
to make the driver aware when exceeding the speed limit of the road. Unfortunately,
the oldest cars will remain without that system.
The aim of this project is to contribute to this research by integrating into a realistic
driving simulator a system able to detect and recognise the speed traffic signs of
the road, taking decisions that help the user make the driving process easier and
safer. The simulator used in this study is the CAR Learning to Act (CARLA), an
open-source simulator for autonomous research that mainly consists of two modules:
the CARLA Simulator and the CARLA Python API module. To detect its road
signs, a cutting-edge object-detection algorithm is used: the You Only Look Once
(Yolo) algorithm. Instead of using a sliding window over several locations within an
image, Yolo sees the entire image during the training and testing phases encoding
contextual information about the object classes as well as their appearances. This
characteristic allows it to be extremely fast while an image is being evaluated.
Identifying the speed traffic signs of a road can have a wide range of utilities. In
this project, two applications are exposed: a warning application, to notify the user
that the vehicle speed has overtaken the maximum allowed speed of that road, and
a control application, to reduce the vehicle speed if it exceeds the road speed limit.
Results show that the detection procedure is achieved satisfactorily with a precision
metric value of 0.92. Moreover, the system is tested both on the CPU and on the GPU, making it reproducible in most of the environments. Running it on the
CPU takes a total time of 130 ms, while running it on the GPU, 8 ms are needed
to evaluate the current CARLA scene and determine whether it exists or not any
speed traffic sign.
+
