摘要
重点探究温室果园作业机器人的障碍物检测及运动目标跟踪技术,目标为提升复杂农业环境中机器人自主作业的可靠性及效率。本文向YOLOv5m基准模型引入CBAM注意力机制和WIoU损失函数,实验结果说明改进后模型检测的精确度提升0.7%,召回率提升至1,测试集检测准确率提升1。然后选择改进的YOLOv5m和DeepSort算法跟踪园间行人,并采用Bytetrack作为对比模型,通过主观评价跟踪视频以及运用评价指标定量评判,DeepSort在轨迹连续、ID稳定一致及遮挡处理能力上明显比Bytetrack表现好,证实了它在动态复杂场景下的跟踪优越性。
This paper focuses on the obstacle detection and motion target tracking techniques for greenhouse orchard operating robots,with the goal of improving the reliability and efficiency of autonomous robot operations in complex agricultural environments.In this paper,the CBAM attention mechanism and WIoU loss function are introduced to the YOLOv5m benchmark model,and the experimental results illustrate that the precision of the improved model detection is increased by 0.7%,the recall rate is increased to 1,and the test set detection accuracy is increased by 1.Then,the improved YOLOv5m and DeepSort algorithms are selected to track the pedestrians between the gardens and the Bytetrack is used as a comparison model.By subjectively evaluating the tracking video as well as quantitatively judging using evaluation metrics,DeepSort significantly outperforms Bytetrack in terms of trajectory continuity,ID stability and consistency,and occlusion handling ability,confirming its tracking superiority in dynamic and complex scenes.
作者
杜海莲
王建华
闫瑾
王观田
王阳
DU Hailian;WANG Jianhua;YAN Jin;WANG Guantian;WANG Yang(Department of Mechatronics and Vehicle Engineering,East China Jiaotong University,Nanchang 330013,China;State Key Laboratory of Intelligent Agricultural Power Equipment,Luoyang,471039,China)
出处
《拖拉机与农用运输车》
2025年第6期29-33,63,共6页
Tractor & Farm Transporter
基金
智能农业动力装备全国重点实验室开放课题(SKLIAPE 2024012)。
