HitoMi-Cam: A Shape-Agnostic Person Detection Method Using the Spectral Characteristics of Clothing

The introduction of HitoMi-Cam marks a significant advancement in person detection technology, particularly in scenarios where traditional CNN-based methods falter due to shape dependency. By leveraging the spectral reflectance properties of clothing, HitoMi-Cam achieves a remarkable processing speed of 23.2 frames per second and an average precision of 93.5%, far exceeding the best performance of CNN models at 53.8%. This method not only demonstrates practical viability on resource-constrained edge devices but also maintains a minimal occurrence of false positives, making it suitable for real-time applications. The study positions HitoMi-Cam as a complementary tool rather than a replacement for CNNs, particularly in unpredictable environments, thus broadening the scope of person detection capabilities.