Stereovision-Based Sensor for Intersection Assistance [chapter]

Sergiu Nedevschi, Radu Danescu, Tiberiu Marita, Florin Oniga, Ciprian Pocol, Silviu Bota, Marc Michael Meinecke, Marian Andrzej Obojski
2009 Advanced Microsystems for Automotive Applications 2009  
The intersection scenario imposes radical changes in the physical setup and in the processing algorithms of a stereo sensor. Due to the need for a wider field of view, which comes with distortions and reduced depth accuracy, increased accuracy in calibration and dense stereo reconstruction is required. The stereo matching process has to be performed on rectified images, by a dedicated stereo board, to free processor time for the high-level algorithms. In order to cope with the complex nature of
more » ... the intersection, the proposed solution perceives the environment in two modes: a structured approach, for the scenarios where the road geometry is estimated from lane delimiters, and an unstructured approach, where the road geometry is estimated from elevation maps. The structured mode provides the parameters of the lane, and the position, size, speed and class of the static and dynamic objects, while the unstructured mode provides an occupancy grid having the cells labeled as free space, obstacle areas, curbs and isles. Introduction Intersections are the most complex traffic situations, as they can be both confusing and dangerous. The standard vehicle trajectories and the standard road geometries that are covered by most of the driving assistance systems do not apply in most of the intersections, but, on the other hand, most of the traffic accidents happen there. Due to the demanding nature of the scenario, the research community is increasingly focused on solving the perception and acting problems related to the intersection. The European research project INTERSAFE, part of the broader project PReVENT (http://prevent.ertico.webhouse.net/en/prevent_subprojects/intersection_safety/intersafe/), had the goal to develop a system that was able to perceive the relative vehicle localisation, path prediction of other objects and communication with traffic lights, in order to warn the driver and simulate active measures. A new joint research project, INTERSAFE-2 (www.intersafe-2.eu), aims at developing a system for cooperative sensor data fusion, based on state of the art passive and active on-board sensors, navigation maps, information from other traffic participants and from intelligent infrastructure, in order to generate a complete and accurate description of the complex intersection environment. The dense stereovision sensor is maybe the sensor that provides the highest amount of usable information, as it combines the visual data with the dense 3D information that can be deduced through precise inference from the binocular view. A reliable stereovision sensor for urban driving assistance has been developed by our team [1]. However, the intersection scenario has some specific demands from a stereo sensor, demands that impose changes in the physical setup and in the software algorithms. A wide field of view is essential for detecting and monitoring the relevant objects at an intersection, but that means lower focal length and consequently reduced working depth for stereo algorithms, and increased lens distortions in the images. The calibration algorithms, combined with an accurate image rectification step, must ensure that the quality of the reconstruction is not affected. The stereovision correlation process has to be performed on rectified images, by a dedicated stereo board to free processor time for the high-level algorithms. Even there are some methods reported in the literature for dense stereo systems calibration [2], [3], their accuracy is not well assessed. Therefore, a calibration method derived from previous approaches used for high precision and far range stereovision is proposed [4], [5], [6] . Camera parameters are estimated on the full-size images. For the extrinsic parameters calibration lenses distortion correction is performed in advance. This ensures the highest possible accuracy of the estimated parameters (which depends on the accuracy of the detected calibration features). The fullsize images are further rectified, un-distorted and down sampled in a single step using reverse mapping and bilinear interpolation [7] in order to minimize the noise introduced by the digital rectification and image correction. Due to the complex nature of the intersection, a solution for perceiving the environment in two modes is proposed: a structured approach, for the scenarios where the road geometry is estimated from lane delimiters, and an unstructured approach, where the road geometry is estimated from elevation maps.
doi:10.1007/978-3-642-00745-3_10 fatcat:6tyilwjxwbeohjewb3bg2j7lxu