This study employed an experimental approach to investigate the effects of various filling degrees of the granular bed and the rotation speed on size-induced particle segregation behavior in an almost fully filled double-walled rotating drum. The distributions of the concentrations of bicolor particles in the system were obtained via image processing to determine the segregation state. The dynamics of the mixture flow including the statistical distribution, the azimuthal angle and radial direction granular temperature profiles were measured and calculated via particle-tracking velocimetry (PTV). The results were similar for the Brazil-nut effect and its reverse in the radial direction at high or low rotation speeds. When the rotation speed or filling degree was changed, the effect on two different particle sizes in the binary mixture system was inconsistent, and different flow mechanisms led to different segregation patterns. Two phase diagrams with the rotation speed and filling degree and the magnitude and position of the maximum granular temperature as the space enabled the prediction of segregation states.