Efficient Real-time Object Detection on Edge NPU Device with TOPST AI

While significant advancements have been made in the field of object detection, state-of-the-art models typically require extensive computational resources. This poses challenges for real-world deployment on resource-constrained edge devices. This study aims to address this issue by exploring the effectiveness of lightweight object detection models paired with a low-power edge device equipped with a Neural Processing Unit (NPU). We examine three popular models - Scaled-YOLOv4, YOLOv5, and YOLOv7 - on the target TOPST AI edge NPU board, which is designed for efficient vision processing. Experiment results using the COCO object detection dataset demonstrate that real-time and accurate object detection on an NPU-equipped edge device is achievable, highlighting the potential viability of this approach for real-world applications.