摘要
随着智能科学的发展,集装箱门式起重机的自动吊运技术成为了近年的研究热点,其中,吊具的自动着箱是关键难点环节。为了实现吊具与集装箱的精准对位,提出了一种基于CenterNet的锁孔关键点视觉跟踪方法。具体来说,首先使用CenterNet进行关键目标检测,然后根据目标的先验几何约束进行筛选,最后对视频流中的关键点坐标进行卡尔曼滤波,实现平滑准确的跟踪。使用1:20集装箱模型制作锁孔数据集,实验结果显示CenterNet相比于其他目标检测网络有更好的泛化能力,并且滤波之后可以以较高的处理速率对关键点进行平滑跟踪,可满足自动着箱操作的要求。
With the development of intelligent technology, the automatic motion control of container gantry crane has become a research hotspot in recent years. Particularly, it is very crucial to realize the automatic grab of container with the spreader. To accurately align the spreader and the container, a CenterNet based visual detection algorithm is proposed to identify the key points of the container keyhole. To be more specific, the targets are first detected by CenterNet, and then screened according to the prior geometric constraints. Finally,the coordinates of the key points in the video stream are filtered by Kalman filter to achieve smooth and accurate tracking. The 1:20 container model is used to make the keyhole data set. The experiment results demonstrate that CenterNet has better generalization ability than other target detection networks. After Kalman filtering, the algorithm can realize smooth tracking of key points with a high processing rate, which meets the requirements of automatic grabbing of container.
作者
郝运嵩
卢彪
刘峰
曹海昕
林静正
方勇纯
HAO Yun-song;LU Biao;LIU Feng;CAO Hai-xin;LIN Jing-zheng;FANG Yong-chun(Institute of Robotics and Automatic Information Systems,College of Artificial Intelligence,Nankai University,Tianjin 300350,China;Tianjin Port(Group)Co.,Ltd.,Tianjin 300461,China)
出处
《控制工程》
CSCD
北大核心
2021年第11期2108-2113,共6页
Control Engineering of China
基金
国家重点研发计划(2018YFB1309000)
国家自然科学基金面上项目(61873132)
广东省机器人与智能系统重点实验室开放基金资助。
