Research On Dynamic Positioning System Of Bolt Machine Based On Vision Technology

Bolting machine is an indispensable equipment in railway maintenance,its accuracy and efficiency have a vital impact on ensuring the good condition of bolts.However,the traditional bolt machine has the problem of inaccurate positioning in the process of use,which needs manual adjustment,and the work efficiency is low.Therefore,the development of a dynamic positioning system based on computer vision technology has become an urgent need.This study aims to propose a dynamic positioning system of bolt machine based on improved camera calibration technology and lightweight YOLOv4 algorithm,so as to effectively solve the positioning problem of bolt machine,improve production efficiency and product quality,and reduce manufacturing cost.Accurate camera calibration is the key to achieve visual positioning.Therefore,this paper will deeply study camera calibration methods,including camera imaging model,coordinate system and coordinate representation and transformation.In addition,the YOLOv4 image processing algorithm will be discussed,and some pre-knowledge must be understood in the processing process,such as neuron model,activation function,neural network and model optimization will be introduced.By using pinhole imaging technique,the homography matrix can be constructed,and the image distortion model can be applied to meet the constraints of the standard value and verticality of the rotation vector,and only a small amount of checkerboard shape images can achieve the desired effect.By introducing tangential distortion and optimized genetic algorithm,the performance of Zhang Zhengyou calibration model is greatly improved,so as to obtain more accurate and reliable measurement data.YOLOv4 is a new lightweight object detection technology,which has a simple architecture,fewer parameters,fewer training requirements and fewer transmission frames,and can be widely used in the industrial field.After the improvement and optimization of YOLOv4,Shuffle Netv2 can be used to replace its main architecture,and SENet module can be incorporated into it to reduce the complexity of calculation.Meanwhile,Swish activation function is introduced to further improve the convergence function of the model.At the same time,a new weighted bidirectional feature pyramid architecture is adopted to further improve the accuracy of feature mixing,and a series of ablation tests are used to evaluate the importance of each channel,and the complexity of the model is effectively reduced by removing excess branches.Based on the proposed improved camera calibration method and lightweight YOLOv4 algorithm,a new dynamic positioning system for bolting machine is designed.After the improved camera calibration method is used for camera calibration,more accurate camera parameters can be obtained,and lightweight YOLOv4 algorithm is used for bolt identification and positioning.In the follow-up experiment,the designed system is tested and verified,and the performance of the system is further optimized through the experimental results.The experimental results show that the dynamic positioning system of the bolt-working machine proposed in this paper can realize efficient and accurate bolt-locating,run at a fast speed,and have good robustness and stability.Based on improved camera calibration technology and YOLOv4 algorithm as the core,combined with lightweight design,an efficient and accurate dynamic positioning system for bolt machine is successfully proposed in this paper.Future research can be further explored in terms of increasing the types of objects detected and optimizing the accuracy and robustness of the system.