基于改进ORB_SLAM3的奶牛喂奶机器人建图算法研究

打开文本图片集

中图分类号：S821.4 文献标识码：A 文章编号：2095-5553（2026）03-0134-08

Abstract：Withthedevelopmentthelivestock industrytowardsstardizationautomation，inteligent milkingrobots haveemerged，aiming toincreasefeeding eficiency，reducerelianceonmanuallabor，improveeconomic outcomes.Thisstudy presentsthe design aninteligentmilking robot navigationsystem，combining SLAM algorithmobject detection.Initially，a hybrid filter using the ExtendedKalman Filter（EKF）forprediction pre-integration forcorrection was designed toimprove data fusionaccuracy stabity between the IMU camera. Additionally，an ECAatention mechanism was integrated into theYOLOv8s network，with the improved YOLOv8s incorporated into the ORB_SLAM3 system to enhance object detection，image segmentation， eliminatedynamic feature points.Through experimental comparisons on the TUMdataset，the improved algorithm demonstrates significant advantages in terms Absolute Trajectory Error （ ATE ） Relative PoseError（RPE）， the RootMean Square Error（RMSE） in dynamic scenes was reduced with up to a 91.41% ，which effectively enhanced the accuracy stability the system.Lastly，the system was successully deployed ontheROS-based robotic platform，generatingahigh-precisiondensepoint cloud mapsfor navigationwithina simulatedranch environment. This studyers aneficient stablesolutionforlocalizationnavigationinintellgentmilking robots，with promising potential for further optimizations in real-time performance broader application scenarios in the future.

Keywords：cow；milking robot；dynamicfeature point elimination；ROS-based robot；hybrid filter；mapreconstruction

0 引言

随着国家大力支持和鼓励畜牧业向规模化、标准化以及智能化方向发展，越来越多的畜牧企业开始意识到智能化设备的潜力，特别是在提升饲养管理效率、减少人力依赖和提高经济效益方面的优势[1]。（剩余13512字）