Book Description

Extraction of local motion signals is crucial for our survival. Lack of information from local motion signals will significantly reduce our ability to discriminate objects from background, avoid obstacles, and navigate. Despite the apparent effortlessness with which we perceive visual motion, there are indications that the underlying neural computations are complex. Three kinds of local motion signals have been distinguished, based on the kinds of spatiotemporal correlations that generate them: Fourier (F), based on 2-point correlations [1]; non-Fourier (NF), based on 4-point correlations [2]; and glider (G), based on 3-point correlations [3]. G signals have two subtypes, expansion and contraction, associated with objects that are looming and receding, respectively. Detection of isolated G and NF signals cannot be mediated by a purely multiplicative cross-correlator or a purely quadratic motion energy model. G signals have recently attracted substantial attention, following the demonstration that a wide range of species (human [3], macaque [4, 5], zebrafish [6], dragonfly [5], and fruitfly [7]) respond to them in similar ways suggesting that there are advantages to using these signals in visual tasks. This work expands the above lines of research in several respects. First, our computational work shows that these motion signals appear in natural scenes and characterizes the basic statistical relationships between them [8]. Second, we report neurophysiological recordings in two distinct visual-speciaist species (macaques and dragonflies) that demonstrate that at the neuronal level, cells response in a similar manner to motion signals in many respects, although there are subtle differences in responses between the species. This convergence at the algorithmic and neural-implementation levels indicate the fundamental biological importance of using the many kinds of motion signals to guide behavior. Finally, we carried out a psychophysical experiment to probe human ability to use multiple kinds of local motion signals simultaneously to solve simple directional task. We found that humans can combine different kinds of motion signals to solve this task, and, interestingly, that sensitivity to different kinds of motion signals is context-dependent.