A Thermal Imaging System with Object Detection Based on High-Precision Temperature Calibration via Thermal Camera Structural Design

This study presents an infrared thermal imaging measurement system equipped with object detection and automatic temperature calibration capabilities. The system is built on a FLIR Lepton 3.5 thermal camera module and a Raspberry Pi 4 embedded platform, integrating 3D spatial enclosure design and DS18B20 digital temperature sensors. A high-precision calibration mechanism is implemented based on the structural design of the thermal imager. Within the enclosure, two sets of reference temperature sensors are embedded: one heated and one at ambient conditions. By applying an interpolation-based correction model, the system accurately calibrates raw thermal data, reducing temperature error from ±5 °C to within ±0.3 °C. The system prioritizes structural thermal stability, modularity, and adaptability across diverse environments, thereby enhancing both the practicality and portability of the calibration approach. Furthermore, a lightweight object detection model optimized for thermal imagery is integrated into the system, enabling precise identification and real-time temperature annotation of pedestrians, electronic components, and industrial equipment in various inspection and biomedical screening scenarios. Experimental results show that the improved model achieves 93.1% mAP, an F1 score of 0.87, and an IoU of 66%, along with a 10% increase in inference speed over the baseline model. These findings highlight that high-precision temperature measurement and multi-functional detection can be realized using cost-effective hardware, delivering performance on par with high-end medical-grade systems and underscoring the system’s potential for intelligent industrial monitoring and healthcare applications.