Viewpoint Planning for Active 3-D Reconstruction of Freeform Surface Parts Based on DDPG-3DCNN

Three-dimensional (3-D) measurement represents a promising approach for assessing the quality of freeform surface parts with stringent precision requirements, where viewpoint planning assumes a critical role in 3-D reconstruction. Prevailing viewpoint planning methods typically operate within discrete spaces, which can result in the loss of critical details due to quantization errors, particularly when applied to products with intricate surface structures. Moreover, conventional approaches encounter difficulties in controlling the overlap rate of the point cloud due to the unique characteristics of freeform surface structures, resulting in diminished registration accuracy of the point cloud. This article introduces a reinforcement learning framework for the viewpoint planning of freeform surface parts with high precision requirements, utilizing the Deep Deterministic Policy Gradient algorithm with 3-D convolutional structures to address optimal viewpoints in a continuous space. Effective control over the point cloud overlap rate is achieved through the design of a unique reward function. Furthermore, constraints related to the eye-in-hand robot measurement equipment are formulated. Through real experiments, our method demonstrates a substantial improvement in point cloud overlap rate control accuracy, surpassing existing methods by an order of magnitude. Our approach achieves higher coverage using fewer viewpoints compared to other reinforcement learning-based viewpoint planning methods. In addition, it exhibits generalization capabilities and can assess models that have not been previously encountered.