针对车辆自主定位实时准确的要求,提出一种结合光流法的车辆运动估计优化方法.采用改进的Lucas-Kanade算法跟踪FAST特征点计算其光流;进而对图像间偏移量进行坐标系转换,获得初始坐标系下车辆的运动估计值;基于偏移量与旋转角度误差服从正态分布的假设,优化更新采用光流法的车辆运动结果,最终映射到世界坐标系中获得车辆运行轨迹.通过测试多组不同车辆行驶轨迹,结果表明:该优化方法突出了光流法的实时性并且克服了其精度差的缺点,有效解决了由累积误差引起的轨迹漂移情况,能够提供车辆准确实时的定位输出.相较于基于特征点匹配的车辆定位其计算时间短,与常用的光流法比较,轨迹更加精确、光滑.
For the requirement of vehicle real-time precise self-localization, an optical flow-based optimization method of vehicle motion estimation has been proposed. The modified Lucas-Kanade was used to track FAST feature points for calculating their optical flow. The coordinate systems of offsets between images are transformed to obtain the estimated value of vehicle motion in the initial coordinates. Based on the assumption of the offset and rotation angle errors obeying normal distribution, the result of using optical flow method was optimized to obtain the vehicle running trajectory when it was mapped to the world coordinate system. Through testing many different vehicle trajectories, the experiment results demonstrate that the optimization method highlights the advantage of real-time performance of optical flow and overcomes the shortcoming of its poor positioning accuracy, and effectively solve the situation that cumulative error makes vehicle trajectory drifted. So it can provide the accurate real-time positioning result. Meanwhile, it has shorter computing time compared with vehicle positioning based on feature points matching, and the obtained trajectory of the study was more accurate and smoother than common optical flow method.