多车互联下的车载视觉感知增强技术研究

Accurate, reliable and complete perception of vehicle’s surrounding environment is essential for high level intelligent driving. Thanks to the development of computer vision technology, the visual perception system installed at one single vehicle is becoming more and more powerful, however, it is still accused of being unreliable and short range, especially when the camera sights are blocked by other vehicles. In order to enhance vehicle’s visual perception ability in complex urban driving condition, this project will study the cooperative visual perception framework between connected vehicles by using the distributed perception and multi-sensor fusion theory. This project will develop a cooperative perception system by connecting multi individual cameras in different cars, and then analyzing visual information from different views to reconstruct vehicle’s 3D environment and eliminate the blind spot..The research in this project can be concluded as follows: 1)wide field 3D reconstruction of driving environment through the study of stereo vison and multi vision technology; 2)high accuracy localization through the incorporation of digital map and visual perception information, for the purpose of calibrating the relative position between different cameras; 3) the mechanism of auto-networking between moving vehicles to form a cooperative perception system that could cover the required perception range; 4)scene cognition and target tracking method through the use of multi-view perception information, achieving the goal of eliminating blind spot and providing robust tracking.

精确、可靠、无盲区的车辆环境感知能力是实现智能汽车高级无人驾驶的重要前提。目前，单车感知能力在不断提升，然而仍无法解决在实际行驶场景中车载传感器因视线遮挡、视野有限所产生的感知缺陷。本项目将以突破单车感知缺陷为目标，基于分布式感知及多源信息融合理论，研究车联网下的多车联合视觉感知方法，建立多车多视觉协同感知机制，融合多车多视角视觉感知信息，实现无盲区超视距感知。.研究内容有：基于机器视觉原理，研究单车多目视觉三维空间建模理论及方法，实现车辆对驾驶场景的动态环视连续建模；结合高精度地图研究多视觉系统空间位置自主式标定技术，精准获取不同视觉系统间相对位置关系；研究联网车辆随机变动场景下多车协同感知网络自组织构建方法，探索多车感知信息融合理论，实现视野扩增与盲区消除；研究以多视角感知信息为输入的多层次场景认知及目标跟踪方法，实现动态目标无盲区识别与接力式连续跟踪，满足智能驾驶汽车无人驾驶需求。

精确、可靠、无盲区的车辆环境感知能力是实现智能汽车高级无人驾驶的重要前提。目前，单车感知能力不足，无法解决在实际行驶场景中车载传感器因视线遮挡、视野有限所产生的感知缺陷，多车互联下的车载视觉感知增强技术的研究是实现智能汽车高级无人驾驶的重要前提。.本项目提出了面向多车感知信息融合的单车行驶环境三维空间建模理论及方法，建立了静态道路环境以及动态目标的空间模型，实现了单车行驶环境的三维重建与交通参与者的意图识别；研究了多视觉系统空间方位高精度获取方法，实现了低成本传感器的厘米级高精度定位；提出了基于多车多视觉系统协同的行车空间感知增强技术，进行无盲区三维量化空间建模，构建了多车协同的动态目标无盲区识别与连续跟踪模型及方法，实现动态目标无盲区识别与接力式连续跟踪；搭建三维环视立体视觉感知系统平台，对所提出的理论和方法进行定性定量评价与验证，取得较好的定性定量验证结果。.项目相关研究共发表论文15篇，其中SCI论文9篇，EI论文6篇，申请发明专利11项，项目核心技术成果获得IEEE国际竞赛冠军，同时获得教育部高等学校科学研究优秀成果奖一等奖与日内瓦国际发明展银奖。本项目研究为高级别自动驾驶技术的落地应用提供了重要保障与支撑。