dc.description.abstract | Search and rescue largely relies on the efforts of the search-and-rescue teams, who take high risks into disasters to rescue and search for victims. Advanced devices in technologies, such as optics, thermal, and acoustic sensors, have been applied in this field for search-and-rescue teams to operate more efficiently and safely. However, some devices are difficult to work in certain environments. For example, thermal and optical sensors cannot be fully functional in fire disaster due to high temperature and thick smoke. Optical sensors are hard to detect in mountain forests. From the two environments mentioned above, the advantages of the acoustic sensing stand out. Therefore, this thesis uses the concept of the sound source localization to develop search-and-rescue devices. In this research, we used Matlab simulation to find out the optimal solution of microphone array geometry for the advantages of small size and omnidirectional positioning. On this basis, we designed a three-dimensional diamond array, and combined it with two sound source localization algorithms, which are Diagonal Unloading Beamformer with Novel Norm Transform (DU-NORT), and Steered Response Power Phase Transform (SRP-PHAT), in two systems for fire and mountain disasters, respectively. In our experiments, we built an acoustic environment simulation to evaluate the robustness of reverberation. Also, we designed two realistic environments of fire and mountain disasters, and added noises of fire alarm and strong winds, respectively. Furthermore, we evaluated the performance of the systems combining the three-dimensional diamond array and sound source localization algorithms. The experimental results showed that using SRP-PHAT algorithm with Adaptive Sound Detector (ASD) in fire disaster can achieve only 10° localization error at the lowest signal-to-noise ratio (SNR) of -10dB, and can be applied in the indoor space of reverberation environment, such as classrooms. Using DU-NORT algorithm with noise robustness system called Mask in mountain disaster can achieve only 22° localization error at SNRs between -25 and -10dB. It demonstrated our systems can effectively assist search-and-rescue operations under the interference of thick smoke and fire alarm in fire disaster, or under the influence of strong winds in mountain disaster. | en_US |