| Abstract [eng] |
This research investigates the application and comparative effectiveness of classical and deep learning methods in developing computer vision systems for resource-constrained devices such as microcontrollers. By employing a range of microcontroller platforms - Arduino Nano 33 BLE Sense, STM32, Arduino Uno, and Raspberry Pi 4 - this study evaluates the performance of object detection, color detection, and MNIST digit recognition tasks utilizing machine learning and image processing methods namely Tensorflow, EloquentTinyML, and OpenCV. Further, an experiment incorporating gesture-based LED control using the MediaPipe and Arduino showcases potential applications for human-computer interaction. This study examines the feasibility of using deep learning and traditional computer vision techniques in IoT applications. |
