Digital Image Correlation Analysis
Digital image correlation (DIC) is a technique for measuring the deformation (i.e. strain) of materials using a sequence of images taken over the course of a mechanical experiment. DIC analysis generally follows 4 steps: capturing an image sequence, tracking features, deformation analysis, and data mining.

Problem: Image-Centric DIC Limitations
Our research group used DIC as our primary tool for non-contact measurement of material deformation. However, our work was inhibited by several limitations of our existing DIC analysis software.
- Computationally inefficient and error-prone motion tracking
- No measurement validation or error estimation
- Poor record-keeping of the steps of completed DIC analyses
- A single type of strain result which required long computation times and large amounts of disk space
- Limited data mining potential due to strain analysis, static reference states, and lack of error estimation
However, the biggest limitation was that our old DIC software was image-centric. From beginning to end, the data was formatted, processed, and saved in arrays matching pixels in the original image. This approach ignores the fact that the discrete feature tracking and the deformation gradient calculation trade location accuracy for displacement and deformation information while introducing and propagating error. Handling the data at the resolution of the image wastes computing resources while misrepresenting the accuracy of the analysis.
Solution: Deformation-Centric DIC
I started by retrofitting our old DIC software with new features and add-ons to compensate for its shortcomings.
- Improved motion-tracking based on the Lucas-Kanade optical flow function set from the OpenCV library
- A “rigid body motion” test for motion tracking validation, error estimation, and artifact detection
- A robust documentation policy that saves every step of the analysis human-and-machine-readable directory tree, eg: .\\images\motion tracking\deformation\data mining
- Deformation gradient-focused data structures which save computational resources and enable a wider variety of data mining strategies without sacrificing accuracy
- Flexible reference states, making it possible to target sections of experiments and compute deformation rates
In addition to these improvements, my DIC software was built fundamentally differently from our old DIC platform. Rather than handling data in an image-centric manner, I implemented a deformation-centric scheme. My DIC software treats deformation data in similar to the nodes and elements found in finite element analysis. This approach facilitates error estimation and enables faster computation times and more efficient use of memory.
Impact: Unprecedented Access to Material Deformation
My analysis suite has become the primary tool of my former research group at Georgia Tech. They are now able to distinguish real deformation from artifacts, validate their measurements, and perform sophisticated strain analyses which are not possible with conventional DIC analysis. For my dissertation, I combined deformation rate analyses with in-situ stress measurements, creating create the first-ever real-time deformation energy measurement based entirely on empirical measurement without simulation or model. By building a DIC analysis suite from the ground up with deformation analysis in mind, I granted my research group significantly more powerful tools for mining mechanical testing data.