基于混合并行计算架构的三维测量与重建

3D measurement and reconstruction on hybrid parallel computing architecture

随着工业应用对物体三维信息需求的增长,3D光学传感器在工业自动化领域得到越来越广泛的应用,然而设计并实现一种高效率、高精度、低功耗、低成本、部署灵活的嵌入式结构光三维成像设备依旧是一个很大的挑战。根据结构光测量系统的组成与特点,设计了一种面向结构光测量系统的混合并行计算架构,并以此架构为基础实现了一种基于多频外差测量原理的结构光三维测量与重建系统。根据所设计的混合并行计算架构的硬件特点,从计算过程和数据类型两个方面对结构光测量算法的计算过程进行了有针对性的优化。最后的实验结果表明,对于一组(1 200×1 600)相移图像计算耗时23.51 ms左右,对测量平板拟合标准差为0.419 mm,所设计并实现的结构光三维测量与重建系统能获得与传统PC平台相当的测量精度,但计算所需能耗比和时间都优于传统PC平台。

With the increasing demands for 3 D information of objects in industrial applications, 3 D optical sensors are paid more attentions and widely in the field of industrial automation. However, it is still a challenge to design an embedded structured light 3 D measure system, which has high efficiency, high precision, low power consumption, low cost and can be deployed flexibly. By thoroughly considering the composition and characteristics of the structured light system(SLS), a hybrid parallel computing architecture for the SLS is designed in this paper. Then, a complete SLS 3 D measurement and reconstruction system, which is based on the principle of multi-frequency heterodyne algorithm, is implemented on the proposed hybrid parallel computing architecture. More specifically, implementations of each process step and data operations are optimized for hardware characteristics of the designed architecture. Benefit from advantages of the proposed architecture, the experimental results demonstrate that a set of phase-shifted images(1 200×1 600) takes about 23.51 ms to calculate, and the standard deviation of the fit to the measurement plate is 0.419 mm, the SLS 3 D measurement and reconstruction system achieves accurate results which are comparable to other traditional PC-based systems. Meanwhile, the energy efficiency ratio and computational efficiency of the proposed system are improved significantly.