目的针对远距离红外飞机目标检测中存在的由于成像面积小、辐射强度较弱造成无法充分提取目标特征进而影响检测性能的问题,提出一种基于全局—局部上下文自适应加权融合(adaptive weighted fusion of globallocal context,AWFGLC)机制...目的针对远距离红外飞机目标检测中存在的由于成像面积小、辐射强度较弱造成无法充分提取目标特征进而影响检测性能的问题,提出一种基于全局—局部上下文自适应加权融合(adaptive weighted fusion of globallocal context,AWFGLC)机制的红外飞机目标检测算法。方法基于全局—局部上下文自适应加权融合机制,沿着通道维度随机进行划分与重组,将输入特征图切分为两个特征图。一个特征图使用自注意力进行全局上下文建模,建立目标特征与背景特征之间的相关性,突出目标较显著的特征,使得检测算法更好地感知目标的全局特征。对另一特征图进行窗口划分并在每个窗口内进行最大池化和平均池化以突出目标局部特征,随后使用自注意力对池化特征图进行局部上下文建模,建立目标与其周围邻域的相关性,进一步增强目标特征较弱的部分,使得检测算法更好地感知目标的局部特征。根据目标特点,利用可学习参数的自适应加权融合策略将全局上下文和局部上下文特征图进行聚合,得到包含较完整目标信息的特征图,增强检测算法对目标与背景的判别能力。结果将全局—局部上下文自适应加权融合机制引入YOLOv7(you only look once version 7)并对红外飞机目标进行检测,实验结果表明,提出算法在自制和公开红外飞机数据集的mAP50(mean average precision 50)分别达到97.8%、88.7%,mAP50:95分别达到65.7%、61.2%。结论本文所提出的红外飞机检测算法,优于经典的目标检测算法,能够有效实现红外飞机目标检测。展开更多
A calibration method for the five essential parameters is proposed. Using the calibration results, the three dimensional (3D) reconstruction can be performed directly. The five essential parameters include the distanc...A calibration method for the five essential parameters is proposed. Using the calibration results, the three dimensional (3D) reconstruction can be performed directly. The five essential parameters include the distance between the camera and the projector, the distance between the reference plane and the camera, the fundamental frequency of the fringe pattern, the scale factor from the image coordinates to the world coordinate system in X axis direction and that in Y axis direction. The proposed calibration method is implemented and tested in our 3D reconstruction system. The mean calibration error is found to be 0.0215 mm over a volume of 400 mm (H)×300 mm (V)×500 mm (D). The proposed calibration method is accurate and useful for the 3D reconstruction system.展开更多
A high-precision vision detection and measurement system using mobile robot is established for the industry field detection of motorcycle frame hole and its diameter measurement. The robot path planning method is rese...A high-precision vision detection and measurement system using mobile robot is established for the industry field detection of motorcycle frame hole and its diameter measurement. The robot path planning method is researched, and the non-contact measurement method with high precision based on visual digital image edge extraction and hole spatial circle fitting is presented. The Canny operator is used to extract the edge of captured image, the Lagrange interpolation algorithm is utilized to determine the missing image edge points and calculate the centroid, and the least squares fitting method is adopted to fit the image edge points. Experimental results show that the system can be used for the high-precision real-time measurement of hole on motorcycle frame. The absolute standard deviation of the proposed method is 0.026 7 mm. The proposed method can not only improve the measurement speed and precision, but also reduce the measurement error.展开更多
In order to realize the online measurement of lamp dimension,the bulb image dimension measurement based on vision(BIDMV)is proposed.The image of lamp is obtained by camera.After image processing,such as Otsu algorithm...In order to realize the online measurement of lamp dimension,the bulb image dimension measurement based on vision(BIDMV)is proposed.The image of lamp is obtained by camera.After image processing,such as Otsu algorithm,median filter,ellipse fitting and envelope rectangle fitting,the dimension of lamp can be calculated.Based on this method,a non-contact real-time measurement system of the lamp’s dimension is developed.The precision of the proposed method is 0.07 mm,and it can satisfy the tolerance of the National Standard GB15766.1-2008.The experiment results show that the proposed method has a faster measuring speed and a higher precision compared with other measurement methods.展开更多
文摘目的针对远距离红外飞机目标检测中存在的由于成像面积小、辐射强度较弱造成无法充分提取目标特征进而影响检测性能的问题,提出一种基于全局—局部上下文自适应加权融合(adaptive weighted fusion of globallocal context,AWFGLC)机制的红外飞机目标检测算法。方法基于全局—局部上下文自适应加权融合机制,沿着通道维度随机进行划分与重组,将输入特征图切分为两个特征图。一个特征图使用自注意力进行全局上下文建模,建立目标特征与背景特征之间的相关性,突出目标较显著的特征,使得检测算法更好地感知目标的全局特征。对另一特征图进行窗口划分并在每个窗口内进行最大池化和平均池化以突出目标局部特征,随后使用自注意力对池化特征图进行局部上下文建模,建立目标与其周围邻域的相关性,进一步增强目标特征较弱的部分,使得检测算法更好地感知目标的局部特征。根据目标特点,利用可学习参数的自适应加权融合策略将全局上下文和局部上下文特征图进行聚合,得到包含较完整目标信息的特征图,增强检测算法对目标与背景的判别能力。结果将全局—局部上下文自适应加权融合机制引入YOLOv7(you only look once version 7)并对红外飞机目标进行检测,实验结果表明,提出算法在自制和公开红外飞机数据集的mAP50(mean average precision 50)分别达到97.8%、88.7%,mAP50:95分别达到65.7%、61.2%。结论本文所提出的红外飞机检测算法,优于经典的目标检测算法,能够有效实现红外飞机目标检测。
基金supported by the National Natural Science Foundation of China (Nos.60808020 and No. 61078041)the Tianjin Research Program of Application Foundation and Advanced Technology (No.10JCYBJC07200)
文摘A calibration method for the five essential parameters is proposed. Using the calibration results, the three dimensional (3D) reconstruction can be performed directly. The five essential parameters include the distance between the camera and the projector, the distance between the reference plane and the camera, the fundamental frequency of the fringe pattern, the scale factor from the image coordinates to the world coordinate system in X axis direction and that in Y axis direction. The proposed calibration method is implemented and tested in our 3D reconstruction system. The mean calibration error is found to be 0.0215 mm over a volume of 400 mm (H)×300 mm (V)×500 mm (D). The proposed calibration method is accurate and useful for the 3D reconstruction system.
基金supported by the National Natural Science Foundation of China(Nos.60808020 and 61078041)the Tianjin Research Program of Application Foundation and Advanced Technology(No.10JCYBJC07200)
文摘A high-precision vision detection and measurement system using mobile robot is established for the industry field detection of motorcycle frame hole and its diameter measurement. The robot path planning method is researched, and the non-contact measurement method with high precision based on visual digital image edge extraction and hole spatial circle fitting is presented. The Canny operator is used to extract the edge of captured image, the Lagrange interpolation algorithm is utilized to determine the missing image edge points and calculate the centroid, and the least squares fitting method is adopted to fit the image edge points. Experimental results show that the system can be used for the high-precision real-time measurement of hole on motorcycle frame. The absolute standard deviation of the proposed method is 0.026 7 mm. The proposed method can not only improve the measurement speed and precision, but also reduce the measurement error.
基金supported by the National Natural Science Foundation of China(Nos.60808020 and 61078041)the National Science and Technology Support(No.2014BAH03F01)+1 种基金the Tianjin Research Program of Application Foundation and Advanced Technology(No.10JCYBJC07200)the Technology Program of Tianjin Municipal Education Commission(No.20130324)
文摘In order to realize the online measurement of lamp dimension,the bulb image dimension measurement based on vision(BIDMV)is proposed.The image of lamp is obtained by camera.After image processing,such as Otsu algorithm,median filter,ellipse fitting and envelope rectangle fitting,the dimension of lamp can be calculated.Based on this method,a non-contact real-time measurement system of the lamp’s dimension is developed.The precision of the proposed method is 0.07 mm,and it can satisfy the tolerance of the National Standard GB15766.1-2008.The experiment results show that the proposed method has a faster measuring speed and a higher precision compared with other measurement methods.
