Aiming at the difficulty of accurately and quickly segmenting complex stem cell phase contrast microscopy images, this paper proposes an adaptive threshold combined with halo correction cell segmentation method. The method analyzes the distribution characteristics of gray-scale images of mesenchymal stem cells with a new adaptive identification (i.e., Rate of rise, Peak intensity and Maximum rate of rise for a dataset). Meanwhile, it uses the halo correction method to eliminate the halo effect caused by the phase contrast microscope, which can accurately and quickly segment irregular mesenchymal stem cells. In order to prove its practicability, we used the phase contrast microscopic images of bone marrow mesenchymal stem cells at different growth stages, assessing by four characteristic indices: precision, recall, F score and confluence error, which were 0.96, 0.90, 0.93 and 0.07, respectively. The results showed that our method was suitable for automatic determination of cell confluence with high accuracy, providing a critical tool for cell culture.