SSD模型在门式起重机障碍物检测中的应用
|
摘要
针对传统的门式起重机障碍物检测方式与避障手段中易受自然环境、现场条件、后期维护等因素的影响以及功能泛化能力较差的问题,提出了一种基于视觉的SSD模型障碍物检测方法。这种检测方式是一种基于回归方法的深度学习目标检测算法,通过对输入图像进行卷积和池化处理等操作提取特征向量,大大提高了对图片中特征检测准确率。采用VOC数据集中的行人、狗、猫、水杯、自行车图片集加上无障碍轨道图片作为训练集,并且训练过程中结合多尺度图像和多环境背景图像来降低复杂环境对检测的影响。实验结果表明,所提供的方法能够有效地提取本文规定的特征,解决了传统门式起重机障碍物检测方式与避障手段的不足,同时提高了运行过程中的安全性。
The traditional gantry crane obstacle detection methods are vulnerable to the influence of natural environment, field condition, the maintenance and so on, and its function generalization ability is poor. In order to solve the problems, this paper proposed a visual obstacle detection method based on SSD model. This detection method is a kind of deep learning detection algorithm based on regression method. It uses convolution and pooling to process the input image, and extract the feature vectors, which greatly improves the accuracy of feature detection in the images. In this paper, we use pedestrians, dogs, cats, water cups and bicycles in the VOC data set as a training set. Meanwhile a free obstacle track picture is added as a training set. In addition, we use combining multi-scale images and multi environment background images in the training process to reduce the impact of complex environment on detection. The experimental results show that the proposed method can effectively extract the features specified in this paper and overcome the disadvantages of the traditional gantry crane obstacle detection and obstacle avoidance, while the security of the operation process is improved.
The traditional gantry crane obstacle detection methods are vulnerable to the influence of natural environment, field condition, the maintenance and so on, and its function generalization ability is poor. In order to solve the problems, this paper proposed a visual obstacle detection method based on SSD model. This detection method is a kind of deep learning detection algorithm based on regression method. It uses convolution and pooling to process the input image, and extract the feature vectors, which greatly improves the accuracy of feature detection in the images. In this paper, we use pedestrians, dogs, cats, water cups and bicycles in the VOC data set as a training set. Meanwhile a free obstacle track picture is added as a training set. In addition, we use combining multi-scale images and multi environment background images in the training process to reduce the impact of complex environment on detection. The experimental results show that the proposed method can effectively extract the features specified in this paper and overcome the disadvantages of the traditional gantry crane obstacle detection and obstacle avoidance, while the security of the operation process is improved.
引文
[1]苟钦.门式起重机収展现状及趋势分析[J].中国科技博览,2010(1):273-273.
[2]胡晓兵,田昆,徐营利,杨雄.基于有限元法的门式起重机抗震研究[J].机械,2018,45(2):15-18.
[3]李书强.超声渡测距在机象鼻粱擅保护的应用[J].港口科技,2011(2):30-31.
[4]黄张禄.动态GPS定位测量的可靠性分析[C].测绘论坛(江苏省测绘学会,浙江省测绘学会,中国测绘学会),2009:3-5.
[5]陈智文,等.基于数据融合的机器人超声测距系统设计[J].信息技术,2013(3):115-117.
[6]Ruder M,Mohler N,Ahmed F.An obstacle detection system for automated trains[C].Intelligent Vehicles Symposium,Proceedings,IEEE,2003:180-185.
[7]李睿.双目视觉立体匹配算法在铁路异物自动识别中的研究[D].兰州:兰州交通大学,2014.
[8]金炳瑞.基于图像处理的铁路轨道异物入侵的自动识别研究[D].兰州:兰州交通大学,2014.
[9]马珊.铁路扣件识别的研究[D].北京:北京交通大学,2012.
[10]刘文琪.基于深度神经网络的铁路异物检测算法[D].北京:北京交通大学,2016.
[11]GIRSHICK R,DONAHUE J,DARRELL T,et a1.Rich feature hierarchies for accurate object detection and semantic segmentation[C].Washington DC:Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition,IEEE Computer Society,2014:580-587.
[12]HE K,ZHANG X,PEN S,et a1.Spatial pyramid pooling in deep convolutional networks for visual recognition[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,2015,37(9):1904-1916.
[13]LIU Wei,ANGUELOV D,ERHAN D,et a1.SSD:single shot multibox detector[C].European Conferenee on Computer Vision,Amsterdam,The Netherlands:Springer,2016:21-37.
[2]胡晓兵,田昆,徐营利,杨雄.基于有限元法的门式起重机抗震研究[J].机械,2018,45(2):15-18.
[3]李书强.超声渡测距在机象鼻粱擅保护的应用[J].港口科技,2011(2):30-31.
[4]黄张禄.动态GPS定位测量的可靠性分析[C].测绘论坛(江苏省测绘学会,浙江省测绘学会,中国测绘学会),2009:3-5.
[5]陈智文,等.基于数据融合的机器人超声测距系统设计[J].信息技术,2013(3):115-117.
[6]Ruder M,Mohler N,Ahmed F.An obstacle detection system for automated trains[C].Intelligent Vehicles Symposium,Proceedings,IEEE,2003:180-185.
[7]李睿.双目视觉立体匹配算法在铁路异物自动识别中的研究[D].兰州:兰州交通大学,2014.
[8]金炳瑞.基于图像处理的铁路轨道异物入侵的自动识别研究[D].兰州:兰州交通大学,2014.
[9]马珊.铁路扣件识别的研究[D].北京:北京交通大学,2012.
[10]刘文琪.基于深度神经网络的铁路异物检测算法[D].北京:北京交通大学,2016.
[11]GIRSHICK R,DONAHUE J,DARRELL T,et a1.Rich feature hierarchies for accurate object detection and semantic segmentation[C].Washington DC:Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition,IEEE Computer Society,2014:580-587.
[12]HE K,ZHANG X,PEN S,et a1.Spatial pyramid pooling in deep convolutional networks for visual recognition[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,2015,37(9):1904-1916.
[13]LIU Wei,ANGUELOV D,ERHAN D,et a1.SSD:single shot multibox detector[C].European Conferenee on Computer Vision,Amsterdam,The Netherlands:Springer,2016:21-37.