Flexible computation of object motion and depth based on viewing geometry inferred from optic flow

Area of Science:

• Neuroscience
• Computational Vision
• Perception

Background:

• Vision is an active process involving eye and head movements.
• Retinal image motion analysis is complicated by self-motion.
• Existing models of eye movement compensation fail for complex viewing geometries.

Purpose of the Study:

• To investigate how the brain accounts for self-motion during visual perception.
• To develop theoretical predictions for object motion and depth perception based on inferred viewing geometry.
• To explore the neural basis of adaptive visual computations.

Main Methods:

• Developed theoretical predictions for perception of object motion and depth.
• Conducted psychophysical experiments simulating viewing geometries using optic flow.
• Utilized a neural network model trained on similar tasks.

Main Results:

• Demonstrated novel perceptual biases in the absence of physical eye movements.
• Showed that these biases are predicted by the theoretical framework and occur automatically.
• Neural network model exhibited response patterns similar to macaque area MT.

Conclusions:

• The visual system automatically infers viewing geometry from optic flow.
• Image motion components are flexibly attributed to self-motion or scene structure.
• Self-motion perception is vital for computing object motion and depth in dynamic 3D environments.