We present an adaptive phase unwrapping method based on geometric constraints and the gradient field without additional images for high-speed three-dimensional (3D) shape measurement. Specifically, we reconstruct the 3D geometry of moving objects frame by frame. We first create a reference phase map at the depth provided by the former frame. Then we optimize the depth value by validating the continuity of the computed unwrapped phase based on the modulus of the gradient field and recalculate the correct absolute phase map with the optimal depth value. After reconstructing the 3D geometry of the current frame, 3D data are delivered to the next frame. In particular, a geometric constraint-based method is applied in the first frame. Experiment results indicate that our approach, which requires only three phase-shifted fringe patterns per frame, can measure moving objects with high accuracy and robustness. Additionally, several isolated objects can also be measured by our method if they are continuous.