Abstract: The accurate acquisition of the location of trapped persons at various disaster scenes can provide decision support for the formulation of rescue plans and further enhance rescue efficiency. In order to obtain the location of people at disaster scenes more accurately, this paper explores an intelligent model for detect and locate the trapped person based on the analysis and summary of existing methods. The system is developed based on the image recognition can automatically analyses real-time video surveillance footage and recognizes whether a person is in a specific distress posture or trapped, so as to send the trapped location and relevant information to rescue personnel for assistance. After testing, the detection accuracy of the system on the personnel posture test set is 94%, and the detection accuracy on the personnel status test set is 90%.

Key words: trapped person, personnel location, distress posture, emergency rescue