Here, the physical and mathematical model is briefly described first, on which the photogrammetric calibration procedure of our Stereoscopic Tracking Velocimetry (STV) system is based. A new hybrid calibration approach is then introduced, which incorporates the use of artificial neural networks. The concept is to improve the performances of conventional calibration techniques of stereoscopic vision. In order to evaluate the quality of the hybrid calibration approach, calibration error is defined for the use of a camera. Our experimental investigation shows that the accuracy in predicting the object frame coordinates has been improved by 30 percents when the hybrid calibration is employed, as compared with the case when only the previous conventional physical and mathematical model is directly applied. It appears that the new idea of using artificial neural networks together with a physical and mathematical model of a system can improve the overall performance of the system. The hybrid method can also be applicable to other general areas in machine vision.
Convective motion in fluid dynamics and heat transfer is the most important phenomenon to be understood since it can greatly influence the performances of fluid and heat transfer systems in various manners. With the advances of modern technologies, new diagnostics for mapping 3D convective flow is veyr necessary for fundamentals of flow physics. Especially, modern computational modeling has been greatly advanced to demand 3D convective-flow diagnostics in order to verify and tune the methodologies and approaches. Conventional velocimetry is either pointwise or 2D. If available, 3D gross-field velocimetry can allow us unprecedented physical insight as well as the needed data for validation of numerical codes and understanding of funamental flow physics. In an effort to meet the need of 3D flow diagnostics, we have developed stereoscopic tracking velocimetry (STV). STV is based on the simultaneous stereoscopic monitoring of numerous particles dispersed in a carrier fluid. It can thus provide time-sequence velocity maps of an entire flow field. Here we briefly present the methodology of STV and its experimental measurement results of 3D flow fields including the traditional flow involving a free jet and the directional solidification for material processing.
Stereoscopic tracking velocimetry (STV) can be a very efficient diagnostics tool for detecting three-dimensional three-component flows with great experimental freedom and computational processing speed but for a restricted region. To achieve the goal of near-real-time measurement with reasonable measurement accuracy, a particle tracking algorithm has been developed, which is an essential part of STV. The developed particle tracking is based on an optimization approach, hence it is a good candidate to be solved by applying computational neural networks. In this paper, we present the new tracking algorithm and its measurement applications to the material processing involving directional solidification as well as to a pulsating free-jet flow. Preliminary comparison of experimental and numerical results is also presented. We believe that by utilizing the massive parallel-processing power of neural networks for optimization, reliable solutions in the STV application can be obtained for near-real-time data extraction and display.