Proposing YOLOv5 Model on Jetson Nano and Evaluating Map Index

The paper focuses on deploying the YOLOv5 model on Jetson Nano using C++ and evaluating the mean average precision (mAP) index. It includes the deployment process and environment configuration, performance testbed, and assessing mAP to evaluate the model. Besides, we perform to track objects on Jetson Nano by integrating YOLOv5 in C++. The results show that the model achieves a fairly high level of accuracy while evaluated on the visual object classes (VOC) dataset. We achieve real-time image processing from 10 to 12 frames per second (FPS) compared with the traditional TensorRT (4 to 5 frames) on Jetson Nano with limited resources. Results will provide important information for using Jetson Nano in embedded applications that require image processing and object detection.