Abstract
Collision avoidance requires that an observer accurately and continuously track the expansion of an approaching object's image. Based on an investigation of the visual system's dynamic response characteristics for this task, we derived a detection model consisting of a linear bandpass filter and a nonlinear criterion detector (presented at VSS 2004). The filter exhibited a biphasic impulse response. In a subsequent set of experiments this model was placed at risk by investigating the implications of the biphasic response characteristic. Experiments were performed in which a circle (representing the outline of a looming object) was presented on an x−y driven CRT. Its radius was subjected to either a pair of pulsed expansions (“PP Pulse”), or an expansion/contraction (“PC Pulse”). If the model is accurate, then the filter's response to two pulses of the circle's radius presented D sec apart will be the sum of the responses to each individual pulse. For certain values of D the individual responses of a PP pulse will reinforce one another, owing to the biphasic nature of the response, while for others they will cancel one another. Thus a PP pulse should be easier to detect for some values of D and more difficult for others.
Three observers (all having normal vision) were tested at each of four different separation time intervals D. A two interval forced choice experiment was conducted in which the circle radius was subjected to either a PP or a PC pulse in one interval and left unchanged in the other. The observers' task was to detect which interval the circle radius was pulsed in. For each observer, threshold pulse amplitude as a function of separation time was obtained using an adaptive staircase algorithm. The experimental results were consistent with the model's predictions. Additional analysis indicated that criterion detection schemes based on either the peak output of the filter or its maximum peak-to-minimum peak output accurately predicted observer performance.