Dynamic Estimation of Oncoming Vehicle Range and Range Rate: An Assessment of the Human Visual System's Capabilities and Performance

The detection of impending collisions and the subsequent choice and regulation of maneuvers to deal with them are general problems of locomotor control that arise in many situations, both human and non-human. When an object moves towards an observer, the size of the image that it projects onto the retina of the observer’s eyes increases, providing a powerful sensation of motion. Physiological and psychophysical research into this “looming” effect provides strong evidence for the existence of neural “looming detectors” that are used by humans and non-humans alike to detect and respond to oncoming objects. Automotive applications constitute an important context for the study of the visual perception of looming. To date, however, this aspect of the driver’s performance has largely been neglected, and human driver models typically incorporate representations of the visual system that are based upon idealized behavior and in some cases questionable assumptions. In this three part study we begin to address the deficiency by quantifying the visual system’s ability to detect and track an object’s approach, as represented by the rate of change of the angle θ that its image subtends on the retina of the eye. In the first part we tested a long-standing assumption of an absolute threshold in the human’s ability to detect dθ/dt, below which humans are unable to discern that θ is changing (and thus that a collision is imminent). The results provide evidence contradicting the threshold assumption, and indicate instead that the detection task is more accurately described as one of signal detection (detection of the signal dθ/dt in the presence of noise) with no threshold limitation. Collision avoidance requires that an observer accurately and continuously track an approaching object’s distance and closing speed. In the second part of this study we investigated the dynamic response of the visual system to changes in θ, employing both psychophysical and classical frequency response techniques. We found that the visual system exhibits a band-pass characteristic in this task that is well described by a linear, minimum phase, second order transfer function. Further analysis revealed that this aspect of the visual system exhibits a biphasic impulse response, which is the focus for the third part of our study. According to the model, certain pairs of “impulsive” stimuli presented in the proper sequence will reinforce one another, and thus be more easily detected, while others will cancel each other and be less so. This final series of experiments provided evidence consistent with this hypothesis. The shortcomings of human driver models based upon current assumptions are discussed, and the development of improved models based the dynamic response characteristics of the visual system and the principles of signal detection are described. To focus our efforts we have assumed a fairly constrained driving scenario (the “Lead Vehicle Braking” scenario), but these results are applicable to any scenario (automotive or not) in which the observer has an unobstructed view of the approach of an object or stationary obstacle.