Polarization de-multiplexing becomes challenging in Stokes vector-direct detection (SV-DD) systems, especially when the rotation of state of polarization (RSOP) and fiber chromatic dispersion (CD) are induced. To address this issue, an adaptive blind RSOP equalizer based on Stokes space is proposed. After theoretical analysis, it is found that the distribution of transmitted signals in Stokes space keeps a paraboloid. Based on this analysis, an adaptive scheme is proposed to track the symmetry axis of paraboloid for the first time. It is worth noticing that not only CD but also high-order modulation formats could not change the paraboloid distribution. Therefore, our algorithm has strong robustness to CD and good suitability for SV-DD systems with different modulation formats. In our simulation, the conventional Stokes space-based method and the training sequences (TS) method are both evaluated for comparison. The results show that our scheme can track RSOP of 20 Mrad/s and 8 Mrad/s at hard-decision forward error correction threshold in 28 Gbaud 4-level quadrature amplitude modulation (4-QAM) and 16-QAM SV-DD systems after 100 km fiber transmission, respectively. No additional computational complexity is required compared to the conventional Stokes space-based method. Moreover, compared to the TS-based method, there is no need to know the actual channel information, contributing to good flexibility.