Monte Carlo Sampling Based Imminent Collision Detection Algorithm

Abstract

Imminent collision detection is an important problem for automotive safety, and is critical for driving assistance systems and fully autonomous vehicles. Imminent collision detection systems require very low false alarm rate, due to the potential outcome of the detection result. Most current approaches are based on Kalman filter or its variants and tend to have degraded accuracy when the underlying noise models deviate from the Gaussian assumption. In this paper, we present a Monte Carlo sampling based Imminent Collision Detection algorithm (MCICD) to achieve improved accuracy by faithfully modeling the noise distributions of the sensor measurements. We further demonstrate a Monte Carlo sampling framework to perform the FPR/FNR analysis for any given collision detection system, as the criterion to evaluate the performance of different collision detection approaches. Experiments with synthetic data and a laser scanner based prototype system have been conducted to validate our approach.