基于位姿状态的林区道路视觉导航

在森林抚育中,为实现林区车辆的精确自主导航,提出一种基于位姿状态的林区道路视觉导航方法。根据林区道路的特点,采用统计学原理选取合适的色差备选通道,利用二维Otsu原理初步分离道路和背景区域,依据真阳率和假阳率定量评价方法选取2B?G为最佳色差通道,因此采用基于2B?G通道的二维Otsu图像分割方法。分析了道路的左右边缘特征,建立基于边缘不完整的位姿状态获取示意图,将选定的道路区域分块求取中心点,计算出试验平台的位姿信息。同时,引入误差校正模型对位姿计算值进行修正,最后选取纯追踪导航方法进行实车试验。试验结果表明,在机器视觉系统引导下,试验平台在不同状态下均能较好地收敛至期望导航线。当速度为0.5 m/s时,在3种初始状态下进行直线导航试验,其全程平均横向偏差为4.6~5.9 cm,平均航向偏差为3.5°~4.0°,稳态航向偏差为1.9°~2.4°,稳态横向偏差为1.6~2.7 cm,试验平台沿林区期望导航线自主行驶的平均相对误差均值为5.2%,说明导航质量具有较好的鲁棒性,满足林区自主导航的作业要求。

Visual navigation of forest road based on pose state

Currently the transportation modes of forest products and managing supplies have the disadvantages of line solidification,low flexibility and low intelligence,while the continuous improvement of forest tending infrastructure provides conditions for the use of driverless forest transportation equipment.In the study of forest location methods,machine vision can perceive the road details in the navigation path,which has become a useful supplement to global navigation such as GPS.Therefore,this study presented a vision navigation method of forest road based on pose state to achieve the accurate autonomous navigation of forest vehicles in the nurture of the forest.According to the charac?teristics of the forest road,a two?dimensional Otsu image segmentation algorithm based on 2B?G channel was firstly proposed by the following steps:the appropriate color?difference alternative channels were selected by principle of sta?tistics;the road and background areas of each color?difference channel were preliminarily separated by using the prin?ciple of two?dimensional Otsu which was combined with the gray scale distribution of 3×3 neighborhood mean image;and the 2B?G channel was selected as the best color?difference channel according to the quantitative evaluation method of true positive rate and false positive rate.Therefore,this study uesd two?dimensional Otsu image segmentation meth?od based on 2B?G channel.Next,under the analysis of the left and right edges of the road,a sketch map of the pose state acquisition based on the incomplete edge was established.The selected road area was divided into blocks to obtain the center point for calculating the lateral deviation and heading deviation of the test platform.Pose calculation was further modified by introducing the error correction model.Finally,the pure tracking navigation method was se?lected for real vehicle test.The test results showed that under the guidance of machine vision system,the test platform could converge to the desired navigation line in different states.The linear navigation test was carried out under three initial states.When the speed was 0.5 m/s and the sampling frame rate was 5 frames per second,the average lateral deviation,average heading deviation,steady?state heading deviation and the steady?state lateral deviation were 4.6-5.9 cm,3.5°-4.0°,1.9°-2.4°and 1.6-2.7 cm,respectively.The average relative error of the test platform driving autonomously along the desired navigation line of the forest road was 5.2%.The results indicated that the navigation quality had good robustness and met the operation requirements of autonomous navigation in forest areas.