A Color-Based Multispectral Imaging Approach for a Human Detection Camera.

In this study, we propose a color-based multispectral approach using four selected wavelengths (453, 556, 668, and 708 nm) from the visible to near-infrared range to separate clothing from the background. Our goal is to develop a human detection camera that supports real-time processing, particularly under daytime conditions and for common fabrics. While conventional deep learning methods can detect humans accurately, they often require large computational resources and struggle with partially occluded objects. In contrast, we treat clothing detection as a proxy for human detection and construct a lightweight machine learning model (multi-layer perceptron) based on these four wavelengths. Without relying on full spectral data, this method achieves an accuracy of 0.95, precision of 0.97, recall of 0.93, and an F1-score of 0.95. Because our color-driven detection relies on pixel-wise spectral reflectance rather than spatial patterns, it remains computationally efficient. A simple four-band camera configuration could thus facilitate real-time human detection. Potential applications include pedestrian detection in autonomous driving, security surveillance, and disaster victim searches.