Abstract Unlike the existing gesture related research predominantly focusing on gesture recognition (classification), this work explores the feasibility and the potential of mid-air dynamic gesture based user identification through presenting an efficient bidirectional GRU (Gated Recurrent Unit) network. From the perspective of the feature analysis from the Bi-GRU network used for different recognition tasks, we make a detailed investigation on the correlation and the difference between the gesture type features and the gesture user identity characteristics. During this process, two unsupervised feature representation methods – PCA and hash ITQ (Iterative Quantization) are fully used to perform feature reduction and feature binary coding. Experiments and analysis based on our dynamic gesture data set (60 individuals) exemplify the effectiveness of the proposed mid-air dynamic gesture based user identification approach and clearly reveal the relationship between the gesture type features and the gesture user identity characteristics.