When a closed space such as a factory, a warehouse, a sports arena or some other similar place, is in emergency due to some kind of natural catastrophes such as avalanches, floods, fires or landslides, the rescue personnel puts their lives in risk, while they seek to mitigate and avoid human losses as much as possible. The use of unmanned aerial vehicles UAVs, may decrease the risk mentioned above, but for the intervention of these to be effective, it is necessary to solve before the various problems that the environment presents for its correct functioning. One of these problems is indoor positioning, which has been addressed through various techniques such as Pseudolitos, Ultra-sound, Vision, Magnetic, among others. However, these techniques have the difficulty of requiring an infrastructure or special equipment at the application site, this means that if one of the equipment is affected during the catastrophe, positioning is also affected. To obtain a solution to this problem, we have proposed through this work three indoor positioning methods using the information of AM and FM radio stations, which are available anywhere in the world and do not require any type of intervention or adaptation to be used. The proposed methodologies use deep learning algorithms, the performance of these was compared with that of a positioning algorithm that uses a regressive KNN, which is commonly used for indoor positioning. Finally, the techniques were taken to an embedded system where their response was tested in real time, reaching results for indoor positioning with an error lower than that presented by GPS devices outdoors