自主导航农业车辆的全景视觉同时定位与地图创建

Simutaneous localization and mapping for autonomously-navigating agricultural vehicle based on panoramic vision

为更好地实现在实际农业作业环境下智能农业车辆的自主导航,提出了基于全景视觉的同时定位与地图创建方法(PV-SLAM)。首先对惯性测量单元(IMU)的姿态进行了解算分析,并设计实现了惯性测量单元的硬件电路模块。其次研究建立了农业车辆运动模型和全景视觉系统观测模型。然后将多目全景视觉(PV)和惯性测量单元(IMU)结合,采用扩展卡尔曼滤波(EKF),实现了自主导航农业车辆的PV-SLAM过程,并具体分析阐释了算法实现流程和步骤。试验结果表明,相较传统视觉SLAM算法,本研究提出的PV-SLAM方法,在较少或无固定路标情况下,获取的环境路标数平均增加80.2%,成功率平均提高15.8个百分点,在x和y方向的平均精度分别提高35.3%和37.8%,定位平均精度提高36.2%。PV-SLAM能较准确完整地提取环境路标信息,且对环境固定路标的依赖较小,因此在实际农业路径作业中运行效果较好。

Simutaneous localization and mapping based on panoramic vision（ PV-SLAM） was developed for the au-tonomous navigation of agricultural vehicles. Firstly, the principle of inertial navigation system （ INS） was analyzed to de-sign the hardware circuit module of inertial measurement unit （IMU）. Secondly, the motion model of agricultural vehicle and the observation model of panoramic vision system were established. Thirdly, the panoramic vision （ PV） was combined with IMU and Extended Kalman Filtering （ EKF） to realize the PV-SLAM process for autonomously navigating agricultural vehicles. Compared with traditional visual SLAM （ vSLAM） , the number of landmarks was averagely increased by 80.2%, and the success rate was increased by 15. 8% using PV-SLAM, under the condition of little or no fixed landmarks. The average accuracies on x and y directions were en-hanced by 35.3% and 37.8% respectively using PV-SLAM method. To conclude, PV-SLAM could perform better in agricultural work owing to its more accurate extraction of enviromental landmarks and less dependence on fixed landmarks.