Akademik Veri Yönetim Sistemi
5. International Anatolian Scientific Research Congress, Hakkari, Türkiye, 21 - 23 Temmuz 2023, ss.1001-1010
Autonomous vehicle
technology is developing rapidly to ensure safe driving and reduce traffic
accident rates. The development of driverless vehicles is largely based on
traffic sign detection technology. Accurate perception of the vehicle's
surroundings and compliance with traffic rules are crucial for safe and
efficient autonomous driving. Within the scope of this study, a dataset of 5
different traffic signs consisting of a total of 1500 images was originally
prepared for traffic sign detection.
Using this unique dataset, model training was performed on the YoloV6
algorithm. In order to provide real-time applicability and mobility, the weight
file of the trained model was run on an embedded system, Jetson TX2. In this way, it is demonstrated that a model-trained
algorithm can also be implemented on smaller and lighter devices without the
need for a large computer by running on an embedded system. Experimental results show that the YoloV6
model achieves a frame per second (fps) rate of 0.215 and a detection time of
4.64 seconds when tested in real time on the Jetson TX2. These findings
demonstrate the feasibility of using the trained model in an embedded device,
paving the way for its integration into driverless vehicles for real-time
traffic sign detection and compliance.