水稻行线检测和导航信息提取方法——基于相机位姿  

Rice Rows Detection and Navigation Information Extraction Method——Based on Camera Pose

作　　者：刘汉文 李彦明[1,2] 刘子翔 黄飞 刘成良 Liu Hanwen;Li Yanming;Liu Zixiang;Huang Fei;Liu Chengliang(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;Key Laboratory of Intelligent Agricultural Technology(Yangtze River Delta),Ministry of Agriculture and Rural Affairs,P.R.China,Shanghai 200240,China)

机构地区：[1]上海交通大学机械与动力工程学院,上海200240 [2]农业农村部长三角智慧农业技术重点实验室,上海200240

出　　处：《农机化研究》2024年第12期15-21,共7页Journal of Agricultural Mechanization Research

基　　金：上海市科技兴农项目(2022-02-08-00-12-F01113)。

摘　　要：为了避免除草机器人在自动驾驶除草过程中碾压和损伤水稻苗,针对单目相机行线检测结果难以提取正确的导航信息导致无法保证除草机器人行驶在正确路径上的问题,提出了基于相机位姿的水稻行线检测和导航信息提取方法。在所规定的坐标系下,利用相机位姿建立像素坐标系与世界坐标系对应坐标的映射关系,根据映射关系对图像进行透视变换并选择感兴趣区域;对感兴趣区域进行图像预处理,采用滑窗法进行水稻行识别;根据坐标映射关系,将水稻行识别结果转换为世界坐标系下的导航信息。试验结果表明:本算法可以实现水稻行线检测与导航信息的提取,同时对于行线弯曲、断行、行线连通等典型情况也适用。在Nvidia Jetson TX2平台上,处理1幅图像并提取导航信息的平均用时约为0.5s,水稻行线识别准确率为97%,成功识别所提取导航线的平均误差为39.0909mm,可以满足除草机器人行线跟踪的实时性与准确性要求。The weeding robot is required to avoid crushing or damaging rice seedlings during the automatic weeding process.GNSS navigation method has the recognition problem to rice rows and the adaptability problem to curved rice rows.Machine vision navigation has the problem of being difficult to extract correct navigation information.To solve these problems,this paper proposes a rice rows detection and navigation information extraction method based on camera pose.First establish the mapping relationship between pixel coordinates and world coordinates,and perform perspective transformation on the original image according to this relationship.For the perspective transformed image,a sliding window method was used to detect the rice rows.The navigation information required by the vehicle can be extracted based on the rice rows detection result through the coordinate mapping relationship.The test results show that this algorithm can meet the target requirements,and it is also suitable for typical situations such as curved rows,broken rows,and connected rows.On the Nvidia Jetson TX2 platform,the running time of the algorithm was about 0.5s,the accuracy rate of recognition was 97%,and the average error of navigation information was 39.0909mm.

关 键 词：水稻行 导航信息 除草机器人 相机位姿 机器视觉 透视变换 

分 类 号：S511[农业科学—作物学]