Quantifying Active Safety System Performance at Pedestrian Detection

Abstract

This practice-oriented paper discusses a novel approach to quantifying the quality of pedestrian detection by in-vehicle night-vision technology. The MaxiMin metric of similarity between sets is sufficiently general to be applicable to a broad range of sensor systems and is not limited to infrared systems. The metric is an asymmetric variant of the Hausdorff Distance that incorporates two sets of weights. The first applies a greater weigh to false negatives (misses) than to false positives (false alarms). The second uses pedestrian height as a proxy for proximity and, hence, risk. The discussion of the mathematics of the metric is followed by a pair of illustrative examples and a discussion of how the metric can inform the design and active safety systems.