Probabilistic Modeling of Motion Blur for Time-of-Flight Sensors.

Area of Science:

• Computer Vision
• Image Processing
• Sensor Technology

Background:

• Synthetic motion blur generation is established for 2D images but lacks methods for 3D data like depth maps.
• Existing techniques do not fully capture the complexities of motion blur in three-dimensional (3D) imaging.

Purpose of the Study:

• To generalize a framework for synthetically generating arbitrary linear and radial motion blur in 3D images.
• To accurately model motion blur effects encountered by Time-of-Flight (ToF) sensors.

Main Methods:

• Extended a prior framework to generate motion blur on planes at arbitrary angles to the sensor plane.
• Employed a probabilistic model to predict invalid pixels in depth maps, considering motion path angles and object velocity relative to the ToF sensor.

Main Results:

• Successfully demonstrated the framework's ability to create synthetic radial, linear, and combined radial-linear motion blur.
• Achieved average Boundary F1 (BF) scores of 0.7192 for radial, 0.8778 for linear, and 0.62 for combined motion blur in invalid pixel prediction.

Conclusions:

• The generalized framework accurately simulates real motion blur in 3D depth maps captured by ToF sensors.
• This advancement provides a robust tool for developing and testing deblurring algorithms for 3D imaging applications.