基于MCN的起重吊装指挥手势信号自动识别  

Automatic Recognition of Command Hand Signals of Crane Lifting Based on MCN

作　　者：张淦 周晓洁 郭辰颢 原毅璨 吴迪[1] 郭聖煜 ZHANG Gan;ZHOU Xiaojie;GUO Chenhao;YUAN Yican;WU Di;GUO Shengyu(School of Economics and Management,China University of Geosciences,Wuhan 430074,China;School of Mechanical Engineering and Electronic Information,China University of Geosciences,Wuhan 430074,China)

机构地区：[1]中国地质大学(武汉)经济管理学院,湖北武汉430074 [2]中国地质大学(武汉)机械与电子信息学院,湖北武汉430074

出　　处：《土木工程与管理学报》2022年第6期131-136,共6页Journal of Civil Engineering and Management

基　　金：国家自然科学基金青年项目(71801197);中国地质大学(武汉)省级大学生创新创业训练计划资助项目(S202110491147);中国地质大学(武汉)教学实验室开放基金资助项目(SKJ2021168)。

摘　　要：为了消除起重吊装因指挥手势信号不规范、交流视野被遮挡、人员注意力不集中等产生的安全隐患,利用基于计算机视觉的混合卷积神经网络(MCN)方法,建立人-机交互高风险场景下起重吊装指挥手势信号识别模型,并提出指挥手势信号识别-确认机制。根据国家标准指挥手势信号的运动特点,选择MCN在起重吊装指挥手势信号数据集上进行训练,得到起重吊装指挥手势信号识别模型。起重机驾驶员对比模型识别结果与直接观察结果,判断是否进行相应的操作。选取五种指挥手势信号为例验证。结果表明:模型在多角度多距离下,平均识别准确率为97.13%,模型泛化能力较强。识别速度为36.9 ms,实际应用中平均帧数可达到27.1 fps,满足实时识别的需求。模型具有一定的实用性。In order to eliminate the potential safety hazards caused by the irregular command hand signal of crane lifting,the blocked communication field of vision,and inattentiveness of worker,the mixed convolutional neural network(MCN)method based on computer vision is used to establish a recognition model of command hand signals of crane lifting in high-risk scenario of human-machine interaction,and the recognition-confirmation of command hand signals mechanism is proposed.According to the motion characteristics of command hand signals in the national standard,MCN is selected to train on the data set of command hand signals of crane lifting,and then obtain the recognition model.The crane driver compares the model recognition results with the direct observation results to determine whether to carry out the corresponding operation.Five kinds of command hand signals are selected as examples to verify.The results show that the average recognition accuracy of the model is 97.13%under multi-angle and multi-distance conditions,and the generalization ability of the model is strong.The recognition speed is 36.9 ms,and the average number of frames in practical applications can reach 27.1 fps,meeting the needs of real-time recognition.The model has certain practicability.

关 键 词：起重吊装 指挥手势信号 计算机视觉 动作识别 混合卷积神经网络(MCN) 

分 类 号：TU17[建筑科学—建筑理论] TU714[自动化与计算机技术—计算机应用技术] TP391.4[自动化与计算机技术—计算机科学与技术]