自动驾驶汽车的环境感知技术是当前运载系统智能化领域的研究热点。基于摄像头和激光雷达的组合感知系统已成为识别行人和汽车等物体的主流方案。然而，在夜间等低光照交通环境中，由于低曝光图像上信息质量的下降、激光点云对物体表示形式的动态信息不平衡以及多模态匹配信息的时空一致性不足等问题，基于图像和激光点云的组合感知模型对夜间动态物体的识别能力难以令人满意。.针对以上在当前研究中尚未有效解决的问题，本项目将建立对低曝光图像上物体显著特征的有效识别机制，实现对图像信息在特征域和时域的联合优化；在点云的时域关联和融合基础上，恢复物体空间形状以建立稳定的点云表示形式；揭示长时域中的物体轨迹信息和物体间的相对运动模式对多传感器的识别结果匹配的影响机理，最终提高自动驾驶汽车对夜间动态物体的识别能力。.本项目将在理论研究上深化时空域优化方法与物体识别算法的结合，并推动自动驾驶感知技术的发展和应用。

The environment perception technology of automated vehicles is one of the research focuses on the field of intelligent transportation systems. The vehicle perception system based on a combination of camera and lidar has become a mainstream solution for recognition of objects such as pedestrians and vehicles. However, in traffic environments with low illumination such as in the night, due to the decline in the information quality of low-exposure image, the imbalance of dynamic information in point cloud representation of objects and the space-time inconsistency of multi-modal matching information, the combined perception model based on image and point cloud is not satisfactory in recognizing dynamic objects. .To address above problems which have not been well solved in current studies, in this research project, we are aiming at: establishing a mechanism to effectively recognize discriminative features of an object on low-exposure images, and jointly optimizing image information in both feature domain and time domain; establishing a stable representation of point cloud by spatial shape completion based on associating and merging point clouds; revealing the influence of longtime trajectory information and relative motion patterns of objects on improving the matching between multi-sensor perception results, and ultimately improving the detection ability of dynamic objects at night for automated vehicles. .This project is expected to deepen the theoretical research on combining time-space optimization and object detection algorithms, and promote the development and application of perception technology for automated vehicles.