Visual Processing of Object Velocity and Acceleration
Abstract
This research analyzed human detection of object motion in random motion noise. Results from experiments demonstrated that: (1) a flexible neural network enhances detection of a feature moving in a constant direction or changing direction slowly; (2) the network uses a highly non-linear facilitation, rather than the linear sum of contrast or luminance signals; (3) a stimulus pattern that moves along the motion path is more detectable than one oriented orthogonal to it; (4) stereopsis does not provide much benefit for detecting motion signals in noise; (5) motion along the z-axis (motion in depth) can be masked by static disparity noise; and (6) the motion system can simultaneously encode both the local directions of small features and the global direction of the flow field. A Bayesian model of object motion and surface segmentation is being developed to explain these observations. Future work will explore whether this model of human motion processing can be implemented computationally in VLSI hardware for detecting moving projectiles in the midst of noise. Twelve papers were accepted for publication in referred journals; four chapters were also supported by this grant.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 27, 1998
- Accession Number
- ADA341070
Entities
People
- Suzanne P. Mckee