Unsupervised domain adaptive person re-identification plays an important role in intelligent monitoring, but the domain shift among different datasets brings great challenges to person re-identification. Studies have reported that the pedestrian images captured from the same camera view have same style in continuous time. If this style information is separated from the pedestrian image, the domain shift problem caused by image style difference will be effectively alleviated. In this paper, a low rank prior guided dictionary learning scheme with domain invariant information separation was proposed. Firstly, according to the low rank priori of the style information, style information and pedestrian identity information in the pedestrian image features were separated. Secondly, according to the domain invariance of the pedestrian attributes of the same identity, the relationship between the visual features and the attributes was established to alleviate the impact of domain shift. Finally, self-training strategy was used to adjust the learning parameters. Experimental results show that the proposed method outperforms the traditional unsupervised domain adaptive person re-identification methods and some unsupervised domain adaptive person re-identification methods based on deep learning in many datasets.