CNN-Based Children Counting in Real-World Using Multiple IR-UWB Radars

With the advancement of loT and sensor technology, the ability to detect human presence and count individuals in real-time has become increasingly essential. This is particularly significant in privacy-sensitive areas where traditional vision sensors are not feasible, making the counting of individuals a key aspect of safety. In this context, we propose the people counting method using Impulse Radio-Ultra WideBand (IR-UWB) radar as the most efficient and adaptable solution in real-world environments. While previous research for estimating the number of people with IR-UWB radar has largely been conducted in controlled experimental environments, real-world settings present challenges, such as numerous obstacles and the multi path effect. To address these, our approach involves the use of multiple IR-UWB radars. Furthermore, to validate our methodology, we set up a challenging real-world scenario to count the number of children in restrooms. Since children have a lower Radar Cross Section (RCS) value compared to adults, distinguishing children signals from multipath signals using a single IR-UWB radar presents a significant challenge. In this paper, we propose that visualizes multiple IR-UWB radar signals into single image, counting the number of children in restrooms using Convolutional Neural Network (CNN). Based on our experiments, our approach not only achieves a 95% accuracy rate in categorizing child counts as 'none', 'single', or 'many', but also reaches a 74% accuracy rate when distinguishing counts of 0 to 4 children.