Research On Intelligent Recognition And Positioning System Of Sewing Material Based On Machine Vision

In the sewing industry,various kinds of sewing equipment are required to cooperate during the garment production process for the cumbersome job.In order for the improvement of the automation degree and cost reduction of manpower production in the sewing process,manipulators tend to be applied to the loading and unloading of sewing materials currently.Considering the random postures and positions of the sewing materials on the assembly line,traditional industrial robot hand teaching and pre-programming methods can no longer meet the process requirements.To overcome the workstage-coherence difficulty in the sewing process,this paper proposes using machine vision technology to recognize the type of sewing material and calculate its posture position information.The paper specifically includes the following content:(1)In order to set image data and verify the visual algorithm,the researches involving the selection of positioning mechanism,camera,lens and light source and the visual platform for sewing material identification and positioning were accomplished according to project requirements,and built a set of visual platform suitable for sewing material identification and positioning.At the same time,the conversion relationship between the imaging plane and the world coordinate system is established,along with the Zhang Zhengyou calibration method used to calibrate the camera,which lays the foundation for the data set construction and attitude positioning algorithm.(2)To solve the problem that the traditional recognition algorithm fails to identify the type of sewing material with high accuracy,this paper proposes using the sewing material image data set to train the ResNet34 network model to realize the type of sewing material.Finally,select the best training parameters after comparing the impact of the network model,the number of network layers,sample size and the learning rate on recognition accuracy.Experiments have verified that the recognition accuracy rate reaches 99.7%,and the average recognition time is12 ms,which meets the requirements of sewing material type recognition.(3)Aiming at the problem of the internal noise of the sewing material,this paper reported to in first-order contour moment to suppress the edge of the slight noise based on the contour curve extracted according to the boundary relation of the seal material.The experimental results demonstrated that the proposed method can accurately extract the outer edge contour and the secondary edge contour,which is clearly identified without breaking the line.(4)To solve the time-consuming problem of image feature matching,firstly,the Harris operator is used to extract the feature points on a large scale in this paper,based on which.SIFT is used to build descriptors further to feature matching then.After verification,the number of feature points can be reduced by 8-10 times and the extraction time by 4-5 times in case that the positioning accuracy of this method conforms to that of SIFT.From the perspective of the incoherent process bottleneck in the sewing industry,this paper establishes a machine vision identification and positioning system that combine target recognition based on deep learning with traditional posture positioning algorithm.The system can accurately identify the type of sewing material and quickly obtain its position information,which provides a relatively meaningful reference for realizing the automation and intellectualization of the sewing industry.