Abstract
Keywords Sensor data fusion, video sensor, radar sensor, perception architecture, high dynamic CMOS camera.
Abstract It is an inborn characteristic of sensing in natural and unprepared environment, that there is a remaining uncertainty with respect to the data perceived. This suggests a stochastic description of all relevant aspects like sensor models, error and probability propagation and an explicit representation of uncertainty on all data levels. The handling and representation of uncertainty and accuracy are of vital concern for the generation of reliable semi-autonomous behavior for vehicle components or of warnings in driver assistance systems that are appropriate and not nerving. To meet the high demands of perception in environments with extremely varying weather conditions Siemens VDO develops video and radar sensors for automotive application as well as a scalable perception system, which processes und fuses video data from mono video streams, disparity data from stereo vision, FMCW radar data, and vehicle data (odometer, inertial data). The perception system performs the following steps: object detection, track initialization, tracking and data association. The paper describes the sensor characteristics, the influence of the sensor arrangement on the results of sensor data fusion and all perception steps in detail.