In order to apply deep-learning-based image recognition technology to autonomous vehicles, it is necessary
to use low-power and small-sized dedicated hardware, that is, edge computing. In this study, we developed a
technology to decide the traffic information obtained by using the deep learning model which is ported to the
NVIDIA Jetson AGX Xavier module which is an edge computing platform. We used the image database for
autonomous driving to training the deep learning model, Faster R-CNN and SSD and ported it to Jetson AGX
Xavier for performance comparison. We measured the precision and accuracy of the trained Faster R-CNN and
SSD models and measured how much the FPS works with the Jetson AGX Xavier. Faster R-CNN showed about
2.4 times higher precision than SSD. On the other hand, SSD showed 1.66 times faster FPS than Faster R-CNN.