Bayesian motion estimation accounts for a surprising bias in 3D vision.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Vision

Background:

• Accurate perception of 3D motion is crucial for survival, particularly for avoiding collisions.
• Humans exhibit surprising errors in judging approaching objects, sometimes failing to detect a direct collision course.

Purpose of the Study:

• To investigate the underlying mechanisms of perceptual errors in 3D motion perception.
• To determine if a neural prior favoring slow velocities influences the estimation of motion-in-depth.

Main Methods:

• Development of a Bayesian computational model to simulate 3D motion perception.
• Independent estimation of parameters related to the visual system's velocity prior.
• Experimental validation of the model using human behavioral data on motion sensitivity and bias.

Main Results:

• The brain utilizes a prior that favors slow velocities when estimating 3D motion.
• This slow velocity prior successfully explains observed biases in human perception of approaching objects.
• The model accurately predicts both sensitivity and bias in 3D motion estimation tasks.

Conclusions:

• Perceptual errors in 3D motion are not random but reflect the brain's reliance on prior knowledge.
• The 'slow velocity prior' highlights the significant role of prior probabilities in interpreting environmental motion.
• Understanding these priors is key to comprehending visual perception and potential deficits.