The thesis explores and implements an automated real-time object recognition system based on the NVIDIA Jetson Nano platform.
The purpose of this study is to develop a high-performance and energy-efficient object recognition system that meets real-time requirements while operating under limited hardware resources.
The research analyzes modern methods and algorithms of computer vision, including YOLOv8, Faster R-CNN, and SSD. A review and comparison of hardware platforms such as Raspberry Pi, Jetson Nano, and Google Coral have been conducted. Data augmentation was applied for model training, and the neural network was optimized using ONNX and TensorRT. A software suite was developed for object recognition, achieving high accuracy (mAP 95%) and a processing speed of 15 frames per second.
The practical significance of this work lies in creating a system applicable in security, transportation, industrial quality control, and AI training. The solutions developed can also serve as a foundation for further research in computer vision and algorithm optimization.
The results confirm the effectiveness of modern approaches to object recognition, neural network optimization, and system integration on embedded platforms.
Research advisor: S.Nechai
