Real-Time Object Detection and Recognition in Embedded Systems Using Open-Source Computer Vision Frameworks

Abstract

The rapid advancements in artificial intelligence and computer vision have facilitated substantial progress in real-time object detection. Embedded systems, particularly Raspberry Pi and Nvidia Jetson Nano, present viable platforms for deploying these capabilities in cost-effective and resource-constrained environments. However, these devices are inherently challenged by constrained computational power and memory limitations. This study is dedicated to the design and optimization of lightweight object detection and recognition systems specifically tailored for embedded platforms. Leveraging the open-source frameworks OpenCV and TensorFlow Lite, we implement YOLOv4-tiny and MobileNet-SSD models. To enhance efficiency, advanced optimization techniques such as quantization and pruning are employed, ensuring real-time performance while maintaining high detection accuracy. The study comprehensively evaluates performance metrics, including detection accuracy, inference latency, and resource utilization, across practical applications such as surveillance and robotics. The results illustrate significant improvements in detection speed and reliability, thereby facilitating the development of scalable, energy-efficient embedded solutions. This research contributes to bridging the gap between state-of-the-art object detection models and the computational constraints of embedded hardware, fostering the broader integration of AI-driven solutions in real-world applications.