Multisensor Low-Cost System for Real Time Human Detection and Remote Respiration Monitoring

The most important features of autonomous Search And Rescue robots are abilities to autonomously detect victims and assess their basic vital parameters, such as respiration and heartbeat status, by using their on-board sensors to classify survivors according to their need of medical care. This paper presents a novel sensor composition for autonomous victim detection and non-contact respiration monitoring with SAR robots having limited on-board computational power, using a combination of commercial low-cost components - a visual sensor and Ultra-Wide Band radar. In the proposed method, a pretrained neural network (MobileNet) is used to process camera frames and detect human presence in real-time. Once the victim is localized, the radar is used to perform respiration monitoring. The proposed method is evaluated by building a prototype and performing measurements on volunteers in different positions, clothing and amount of subjects in frame.