Performance Analysis of Low-Cost Wireless AIoT for Real-Time Object Detection: Arowana Monitoring Case Study

บทคัดย่อ

This study implements wireless AIoT (Artificial Intelligence of Things) for monitoring Asian arowana in aquarium environments and investigates the performance of wireless IoT networks on AI inference computation rates in real-time object detection systems. Employing the ESP32-CAM for video capture and the YOLOv3-tiny model for object detection, this research utilizes HTTP over Wi-Fi for communication protocols on the deployed wireless IoT networks. Experimental results indicate that higher throughput transmission does not necessarily enhance AI inference computation rates. Increased bandwidth consumption can lead to inefficient use of radio resources without significantly benefiting AI object detection. Our experiments highlight this issue, emphasizing the need to optimize both wireless IoT networks and AI algorithms to improve the performance of AIoT systems. These findings provide crucial insights into avoiding unnecessary radio resource usage to maintain IoT network effectiveness and mitigating network constraints to enhance AI inference computation rates. This research contributes to the development of practical and cost-effective AIoT solutions for aquaculture and other applications.