Predictive estimation of radioactive sources using Bayesian theory is of great importance in nuclear emergency response. To this aim, we put forward an unscented particle filtering algorithm based on divide-and-conquer sampling. Our method exploits the information acquired by mobile detection robots to search for radioactive sources at unknown locations. Firstly, we assume a circular and extremely thin NaI (TI) scintillator detector, then analyze the geometric relationship between the detector cross-sectional area and the square of the radiation source distance, and finally establish a radiati...