Yutong Zhang¹ and Ang Li², ¹USA, ²California State Polytechnic University, USA

In the realm of home safety, the heightened risk of injury among unsupervised children, particularly from window-related falls, represents a significant challenge [1]. This study introduces an innovative solution to mitigate such risks through a novel integration of technology and artificial intelligence. We propose a comprehensive system that harnesses the power of Deep Learning, specifically utilizing the YOLOv8n algorithm, in conjunction with a Raspberry Pi platform for real-time hazard detection [2]. To address the critical aspect of data transfer, our system employs Firebase for efficient and timely communication between components. Acknowledging the limitations posed by initial model inaccuracies, our approach involved augmenting our dataset to ensure diversity and mitigate the risk of overfitting, thereby enhancing the model's predictive accuracy. This paper details our experimentation with various configurations, including an attempt to utilize YOLOv8x, which was ultimately revised to YOLOv8n due to computational constraints of the Raspberry Pi [3]. The robustness of our system was rigorously tested across diverse scenarios involving windows and doors, establishing a comprehensive dataset that underscores the system's effectiveness. By integrating real-time detection with an intuitive user interface, our system offers a proactive tool for parents to enhance home safety for their children. This contribution not only addresses a pressing societal issue but also advances the application of Deep Learning and IoT technologies in the domain of domestic safety.

YoloV8n, Object detection and recognition, Raspberry PI, OpenCV, Picamera2