The invention discloses an incomplete attribute tagged pedestrian re-identification method and system based on two-level fusion. The method comprises the steps that firstly, on the basis of the premise that attributive features artificially tagged are more precise, by utilizing cross-level sparse reconstruction consistency of low-level features and high-level general attributive features, and artificial tagged attributes are preliminarily fused to study out complete and accurate high-level attributive features; then, an attention and saliency model is provided through attributive attention and saliency based on statistical significance to measure distances between pedestrian attributive vectors; finally, sorting postgenital fusion is conducted on a sorting result based on the low-level features and an attributive feature sorting result based on attention and saliency. According to the incomplete attribute tagged pedestrian re-identification method and system based on the two-level fusion, the validity is verified on a VIPER data set, and meanwhile it is known that the retrieval effect can be greatly promoted by introducing incompletely tagged attributes for pedestrian re-identification.