As a typical mechanical element, the vibration of a rotating disc during actual operation is usually three-dimensional. Existing studies generally focus only on two-dimensional vibration in the plane, ignoring the existence of complex off-plane vibration modes in the depth direction. In order to obtain more comprehensive 3D vibration displacement information, we propose a non-contact measurement method of 3D vibration displacement based on monocular vision. Fig. 1 shows the overall framework of the proposed three-dimensional vibration displacement measurement method. The framework splits the 3D vibration displacement measurement task into two sub-modules, in-plane displacement measurement and off-plane displacement measurement, and integrates the two sub-modules into the same vision measurement system through the coordinate extraction module. Among them, the in-plane displacement measurement module uses a pixel-level semantic segmentation algorithm to predict the segmentation mask of the rotating disk frame by frame. Then it regresses the position changes in the x and y directions at the center of the mask to regress the in-plane displacement of the measured target. The off-plane displacement measurement module uses an advanced monocular depth estimation algorithm to predict the depth information of the corresponding video frame of the two-dimensional scene, so as to extract the displacement change in the z direction at the target position. With the coordinate extraction module we designed, it is possible to integrate the two submodules into the same vision measurement system from the coordinate index of the center point of the semantic segmentation mask to the pixel position of the target detection point corresponding to the depth map. Finally, we can regress the 3D vibration displacement of the rotating disc by integrating two sets of displacement signals, the position of the center point of the semantic segmentation mask, and the offset of the depth value of the depth index point over time. We applied a high-speed camera and an RGB-D camera to create in-plane and off-plane rotating disk datasets. It has been verified that the proposed method achieves excellent 3D vibration displacement measurement performance in the dynamic environment of a rotating disk, utilizing only a single camera as the video data collector.

Rotating disc,3D vibration displacement measurement,Monocular depth estimation,Semantic segmentation,Computer vision technology application

暂无