Design and Implementation of a Near Real-Time Human Detection Robot Using YOLO Framework and IoT Technologies

Abstract

Deploying advanced deep learning models on resource-constrained, battery-powered robots presents significant system-level challenges in balancing performance, power efficiency, and operational reliability. This paper details the design and implementation of a near real-time human detection robot, demonstrating a holistic system architecture for deploying the YOLO (You Only Look Once) framework on a resource-constrained industrial microprocessor. The system synergistically integrates YOLO-based computer vision with a robust embedded architecture—featuring the STM32MP157F microprocessor, SIM800L GSM module, and ESP32 board—to achieve near real-time image processing at 0.5–1 second per image and detection accuracies between 75–80%. While this is not suitable for applications requiring instantaneous reaction, such as autonomous navigation in dynamic environments, this rate is sufficient for the intended security and surveillance scenarios, where the primary objective is to reliably detect and log human presence over time. A key system-level contribution is its multi-modal communication strategy (GSM/Wi-Fi/Bluetooth), ensuring reliable data transmission even with unstable internet connectivity. This enables efficient human identification and tracking across diverse sectors, including security surveillance, search and rescue, elderly care, and industrial safety monitoring. The research provides a validated blueprint for enhancing operational efficiency and public safety, showcasing a pragmatic and adaptable robotic solution for deploying advanced AI under real-world constraints.