AI-Powered Video Monitoring: Assessing the NVIDIA Jetson Orin Devices for Edge Computing Applications
Filipe Pinarello Scalcon, Rondon Tahal, Masoumeh Ahrabi, Yixin Huangfu, Ryan Ahmed, Babak Nahid‐Mobarakeh, Shahram Shirani, Carlos Vidal, Ali Emadi
Abstract
This paper evaluates the performance of the NVIDIA Jetson Orin family of devices for AI and edge computing applications, focusing on a parking lot surveillance example with CVEDIA-RT software. The NVIDIA Jetson Orin AGX Developer Kit is used as a means to emulate the Orin NX and Orin Nano devices. A testing procedure based on augmented scripts is presented to assess key performance indicators like RAM, GPU and CPU usage across the Orin NX, and Nano models. By employing the parking lot footage as a real-world test for intruder detection, it was found that all models consistently deliver at least an average of 10 FPS, with higher-end models outperforming the lower-end Orin Nano device. Additionally, the YOLOv4 algorithm is deployed with DeepStream on the Jetson Orin Nano Developer Kit, showcasing that the 15 W configuration is suitable for surveillance applications, achieving 13 average FPS.