Binocular thermal metalens for gesture recognition

• We demonstrate a practical metalens–AI gesture recognition system using binocular thermal metalenses, fabricated with a simple Pancharatnam–Berry phase design that is compatible with low-cost i-line lithography. • The binocular metalens performs a 98.85% recognition accuracy by combining stereo thermal images with a CNN model. • Our results show that combining metasurface optics with deep learning is effective, providing a robust and low-cost solution for intelligent thermal sensing with potential applications in human–machine interaction, AR/VR wearables, and thermal imaging. This work presents a deep learning assisted thermal gesture recognition system based on three types of metasurface optics: monocular, achromatic, and binocular metalenses. The monocular and binocular designs are polarization-selective and fabricated using a scalable i-line lithography process with a simplified Pancharatnam–Berry phase approach, offering a low-cost yet effective solution for thermal imaging. In recognition tasks, the monocular and achromatic systems achieved accuracies of 92.76% and 94.80%, respectively. Importantly, the proposed binocular metalens, combined with a dual-channel convolutional neural network, achieved an accuracy of 98.85%, demonstrating enhanced robustness to low contrast and partial occlusions. These results highlight the synergistic potential of metasurface design and AI-based post-processing in advancing intelligent thermal sensing, with promising applications in human–machine interaction, AR/VR wearables, and low-cost thermal imaging platforms.