Vehicle Environment Detection by a Combined Low and Mid Level Fusion of a Laser Scanner and Stereo Vision

Abstract

A sensor system consisting of a multi layer laser scanner and a stereo camera is used to observe the surrounding environment of a vehicle. By a novel multi level fusion framework the heterogeneous sensor data can be merged on different processing levels. The following paper gives a short introduction to the processing steps of the sensor data and focuses on the lower raw level and middle object level fusion. Here, the low level fusion obtains a more precise object discrimination, whereas the mid level fusion is able to solve conflicting sensor statements with competing object information. A combination of both, obviously results in a more accurate environmental description. The effectiveness of both fusion approaches is tested and compared with real road reference data, obtained from highly precise GPS data.