Stereo Reconstruction of Droplet Flight Trajectories

Area of Science:

• Fluid Dynamics
• Computer Vision
• Optical Measurement

Background:

• Traditional multi-camera tracking struggles with droplets due to similar appearance.
• Accurate 3D trajectory reconstruction is crucial for understanding droplet dynamics.

Purpose of the Study:

• To develop a robust method for extracting 3D droplet flight trajectories.
• To overcome limitations of conventional tracking in dense droplet scenarios.
• To improve 3D reconstruction accuracy beyond traditional triangulation.

Main Methods:

• Utilized high-speed stereo capture for 3D data acquisition.
• Employed local motion models for individual droplet tracking in 2D.
• Learned a global, non-linear motion model from 2D tracks to infer 3D positions.
• Developed a concomitant solution for 3D trajectory and motion model estimation.

Main Results:

• Successfully reconstructed 3D flight trajectories for hundreds of droplets.
• Demonstrated superior performance compared to 3D reconstruction via triangulation.
• Showcased the ability to estimate 3D locations even when droplets are not visible in all cameras.
• Achieved significant improvements in 3D estimation accuracy using unsynchronized reconstruction.

Conclusions:

• The proposed method effectively extracts 3D droplet trajectories by integrating local and global motion models.
• The learned global motion model enhances reconstruction accuracy and relaxes stereo system simultaneity constraints.
• This technique offers a robust solution for complex droplet tracking problems in fluid dynamics research.