Pedestrian detection with thermopiles using an occupancy grid

Abstract

The increase in road safety has been an essential issue for car manufacturers for the past two decades. While present sensor systems mainly focus on the validated detection of conventional crash partners like other vehicles or poles, there is still a lack for a reliable and cost efficient solution for the detection of pedestrians. In this context, a passive infrared system is a viable alternative mainly due to its affordability and environment-friendliness. A pedestrian detection system based on infrared (IR) sensors is presented. The system consists of an array of thermopile sensors which are arranged on the front bumper of the automobile in a way that the region of interest (ROI) is covered by the sensors field of view (FOV). Since thermopiles do not provide distance information, the sensor arrangement should also facilitate object localization. The optimal sensor arrangement is computed through an optimization algorithm. However, for an accurate position estimation, the sensor arrangement should be supported by suitable probabilistic techniques. For this reason, an approach was developed to map the measurements by means of a sensor model to an occupancy grid. The grid contains a probabilistic value in each cell and is ego-motion compensated. The combination of the optimized sensor arrangement and the occupancy grid technique are discussed here and a reliable method for the position estimation of pedestrians is presented.