A Multi-View Structured Light Measurement Method Based on Pose Estimation Using Deep Learning

To enhance the effectiveness and flexibility of the data alignment in the multi-view measurement system, a measurement strategy based on pose estimation using deep learning is proposed. The object pose is estimated and established through a single-shot pose estimation network. Then, the coarse alignment of the data acquired from different views is performed using the estimated 6D pose. The ICP algorithm is utilized for global refinement. Different shapes are used to verify the effectiveness, robustness, and flexibility of the deep learning-based multi-view measurement strategy. Furthermore, error comparisons of data fusion using markers and deep learning are implemented. The translation error of pose estimation is 1.8-5 mm, and the angle error can reach 0.5-1 degree. The difference between the marker-based and proposed data alignment method is only 0.02 mm. The proposed method can achieve comparable data alignment accuracy with the marker-based method. Moreover, it increases the flexibility and convenience of the data alignment and provides an improved way for existing marker- and shape-based multi-view measurement systems.