解耦光流运动场模型的车载平台仿真

Simulation on Vehicle Platform Based on Decoupled Optical Flow Motion Field Model

针对无人车平台利用光流进行载体位姿估计时面临的不同运动状态光流矢量解耦与分析问题,推导了光流运动场模型,分析了载体在六自由度位姿独立变化时的解耦光流运动场模型;根据解耦光流运动场模型,设计了车载平台的仿真算法,并给出完全解耦的仿真结果;利用解耦光流运动场模型量化分析了仿真结果的正确性;利用KITTI数据集的平移、旋转两个典型场景开展真实光流解耦实验,进行了模型分析、仿真过程、真实数据、对比结果的一致性验证。结果表明:所给出的解耦模型分析、仿真算法、仿真与真实结果以及对比分析不仅可用于车载平台利用光流开展位姿解耦估计中的误差分析和算法验证,还对深入理解光流运动成像、开展无人车平台光流应用的研究具一定的借鉴和指导。

Aiming at the problems of decoupling and analysis of optical flow vectors with various motion states confronted in the pose estimation process of autonomous vehicles(AVs) using optical flow, the optical flow motion field model(OFMFM) is derived and the decoupled optical flow motion filed model(DOFMFM) is analysed as the vehicle poses change independently in six degrees of freedom. According to the DOFMFM, a simulation algorithm is designed for the vehicle platform, and the completely decoupled simulation results are presented. The DOFMFM is applied to quantify and verify the simulation results. Two real scenes of translation and rotation from the KITTI dataset are utilized for the flow-decoupled experiments. The consistency among model analysis, simulation process, real data and comparison results is verified. The results show that the proposed decoupled model analysis, simulation algorithm, simulation and real results together with comparison analysis can not only be applied in the error analysis and algorithm test of pose estimation, but also provide a reference or instruction for the improvement in the understanding of optical flow motion imaging and the research on the optical flow based applications on an AV platform.