Abstract: Aiming at the problem that the current traditional vibration sensor is limited by the installation and the number of measur‐ ing points when measuring the displacement of the rotating body， this paper uses a high-speed industrial camera as the acquisition medium， and collects the rotor vibration video on the rotor vibration test bench. The visual vibration measurement method tracks the full-field vibration displacement of multiple rotor targets. The feature pyramid network structure is introduced into the residual neural network， and the improved feature extraction backbone network is established by combining the attention mechanism. The identity re-identification method is used to strengthen the correlation of target displacement between adjacent frames， and to track the full-field vibration displacement signals of the rotator. Qualitative and quantitative comparisons of different network models on the rotor vibration displacement measurement dataset show that the network model proposed in this paper can obtain a tighter fit when the bounding box is regressed. The collected eddy current displacement signal is used as the standard value to compare the two rotor displacement signals， and the experimental results show that the waveform and spectral noise fitted by the multi-target tracking algorithm in this paper is the smallest and can match the eddy current signal. The experiments also prove the generalization performance of the algorithm in this paper， which reflects the engineering application value of visual measurement in the field of vi‐ bration displacement tracking of rotating bodies.