无人驾驶完成自主行驶任务的前提是车辆能够正确理解交通场景，而道路交通环境的开放性与不可预知性是对无人驾驶环境感知智能的最大挑战。本项目研究计算机环境下的无人驾驶环境感知智能的测试与可靠性评估问题，拟采用原子任务及其多粒度组合测试模型，根据不同感知要素对无人车安全行驶影响的重要程度，研究时-空行驶安全约束的环境感知智能可靠性评估方法；针对极端交通场景数据稀缺的现实困难，研究时空操纵的异常交通元素运动交通场景数据交互生成方法，以及雨雪雾天气效应的对抗式迁移图像生成算法，保证提供能对环境感知智能最大测试覆盖的交通场景数据；所构建的数据测试平台目的是支持无人驾驶环境感知智能的离线测试。通过强化无人驾驶的离线测试能力，可降低道路测试风险成本，提高无人驾驶的测试效率。

The premise for unmanned vehicles to complete autonomous driving tasks is that vehicles can correctly understand traffic scenes. However, the openness and unpredictability of traffic environments is the biggest challenge to the environmental perception of unmanned driving. This project studies the intelligent test and reliability evaluation of the autonomous perception. The atomic task is proposed to be utilized in the basic testing process and combined as multi-granularity test genres. According to the importance of different perception factors for safe driving, the reliability evaluation method of the environmental perception based on the spatio-temporal safety constraints is studied. To overcome the data scarcity of extreme traffic scenes, the interactive generation algorithm of traffic scene data is studied. Moreover, image generation based on spatio-temporal manipulation algorithm for different weather conditions based on adversarial migration is proposed. These approaches provide the traffic scene data with maximum test coverage of environmental perception. The testing platform is designed to support the off-line test of perception intelligence of autonomous driving systems. As a consequence of strengthening the off-line testing ability of unmanned vehicles, the risk of field testing can be reduced. Moreover, the testing efficiency of unmanned vehicles can be improved.