Image and Laser Scanner Processing as Confident Cues for Object Detection in Driving Situations

Abstract

In the area of advanced driver assistance and automation systems knowledge about the vehicle environment is becoming more and more important in order to increase traffic safety. This paper is concerned with the detection and tracking of objects in the proximity of the ego-vehicle while driving on highways. For this purpose, a camera sensor and a laser scanner are used. The processed data of the sensors is then fused at object level in a competitive way. The paper focuses on the generation of object observations by applying the mentioned sensors. In the case of the camera system, an image processing method based on texture information is presented. The texture information is adaptively calculated in order to be independent of the lighting conditions. Taking into account knowledge about the image structure in driving situations, texture segments are classified and object observations are generated. In comparison to other methods, objects are detected independently of any features, model and movement assumptions. For object generation from laser scanner data, a method characterizing detected object contours by means of a shape indicator (long, corner, round, concave etc.) is proposed. Different to other works in this field, in the method presented here explicitly obtains the objects’ optimal reference point and the observability of the objects’ components. The experiments conducted both with simulated and real data show the plausibility of the methods to be used as cues for an object fusion system.