Development of Object Detection Models Compatible with Edge TPU Accelerator for Low-Cost Single-Board Computers

With the rapid advancements in deep learning models over the last decade, artificial intelligence applications such as object detection, object tracking, and audio processing and analysis have become increasingly prevalent in everyday life. These software-based advancements have brought with them the need for high-cost hardware. In this study, low-cost embedded single-board computers, such as Raspberry Pi 4-5 and Jetson Nano, which are frequently used in academia and practice, have been compared in terms of performance while running object detection models, and new solutions have been developed. At the core of these solutions lies the Edge TPU co-processor, designed for lower-performance single-board computers like the Raspberry Pi, where the computational load is handled by the CPU. The object detection models developed for single-board computers have been trained to be compatible with the Edge TPU, and real-time object detection was performed using a camera, with performance metrics being calculated.