Implementation of Victims Detection Framework on Post Disaster Scenario

Disasters are prone to occur in Indonesia due to geographical factors, such as tectonic plate movements, which can cause an earthquake. Earthquakes are one of the most frequent disasters, they have broad impacts in a short time and are unpredictable. Thus, an extensive search process in a short time is highly critical to determine the victims location. In this paper, a victims detection framework is developed starting from acquiring images using an unmanned aerial vehicle and further processing using convolutional neural network (CNN) to locate victims robustly on post-disaster. Input images are then sent to victim detector dedicated ground station server for further high processing robustly locating the possibility of victims. A simulation system mimicking a real environment is developed to test our framework in real time. A transmission protocol is also developed for effectively transmitting data between the robot and the server. The treatment on the detection process of the victim is different from the normal human detection, some pre-processing stages are applied to increase the variation of the given dataset. An embedded system is used for taking images and additional sensors data, such as location and time using Global Navigation Satellite System.