DESIGN AND IMPLEMENTATION OF JETSON NANO FOR TRAFFIC SIGN DETECTION WITH 5G* NETWORK SUPPORT ON A SELF-DRIVING CAR PROTOTYPE

• Main Author: Arifatullaily, Hidayati
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/82260
• Institution: Institut Teknologi Bandung

This research aims to develop and implement a traffic sign detection system on a JetRacer self-driving car prototype using Jetson Nano, supported by 5G network technology. The proposed system leverages AI technology by utilizing deep learning algorithms, namely Single Shot Multibox Detector (SSD) MobileNet-v2 and Convolutional Neural Network (CNN) ResNet-18, to detect and recognize traffic signs in real-time. Although the initial goal of the research included the integration of a 5G network to enhance the system's speed and responsiveness, the implementation of the 5G network was not realized in this experiment. In the design phase, the hardware subsystem consists of the JetRacer AI Kit, which includes Jetson Nano, a camera, motors, servos, and an RC car chassis. Jetson Nano was chosen for its capability to efficiently run AI algorithms. The software subsystem encompasses the programming and configuration of the software for traffic sign detection. SSD MobileNet-v2 and CNN ResNet-18 were implemented to perform the detection and classification of traffic signs from visual data captured by the camera mounted on the JetRacer car. The implementation began with assembling the JetRacer hardware according to the provided manual. The Jetson Nano was mounted on the RC car chassis and equipped with a camera connected to capture images of the road and traffic signs. Once assembly was completed, the software was installed and configured on the Jetson Nano, including the installation and training of object detection models. SSD MobileNet-v2 was used for fast detection with adequate accuracy, while CNN ResNet-18 was used for traffic sign classification with higher accuracy and could be integrated with the JetRacer. From the test results, the SSD MobileNet-v2 algorithm demonstrated effective traffic sign detection capabilities with an accuracy of 59.6% for stop signs, 57.7% for turn_left signs, and 60.2% for turn_right signs, although it did not reach the 70% accuracy threshold. These results indicate that JetRacer can recognize important traffic signs under certain conditions, opening up opportunities for further improvement through more extensive data collection or algorithm adjustments. Additionally, the CNN ResNet-18 testing showed that JetRacer successfully adopted new behaviors, such as stopping when encountering a stop iv sign on track 1 with a training score of 237/364 and an average loss of 0.062. Similarly, on track 2, JetRacer successfully adopted new behaviors with two more complex scenarios. JetRacer successfully detected left or right turn signs when encountering turn_left or turn_right signs and stopped when encountering stop signs, achieving a training score of 458/516 and an average loss of 0.0362. Analysis of the test results indicates that Jetson Nano can efficiently run both algorithms, even without 5G network support. The latency produced is still within acceptable limits for prototype-scale self-driving applications. The main challenge encountered was the inability to integrate the 5G network, due to limitations in infrastructure and available devices during this research. This research concludes that a traffic sign detection system using Jetson Nano with SSD MobileNet-v2 and CNN ResNet-18 algorithms can function well in a simulated environment. However, for broader and more realistic applications, especially those involving real-time communication between vehicles and a central server, 5G network support will be crucial. The failure to integrate the 5G network in this research highlights the need for further research to overcome the existing technical and infrastructural challenges. Overall, this research significantly contributes to the development of autonomous vehicle technology, particularly in traffic sign detection. The results obtained can serve as a foundation for the development of more advanced systems and broader implementation, especially with the full support of 5G network technology in the future