| As an important control and protection device in the power system,high-voltage circuit breakers are widely used in transmission and distribution networks,and their operation,maintenance and repair status are closely related to the stability of the entire power system.As the core component of the circuit breaker,the operating mechanism is also the concentrated part of the circuit breaker failure,and its operation status can be measured and judged through the mechanical characteristic parameters.In view of the current problems of different measurement methods for the mechanical characteristics of circuit breakers and the need to improve the accuracy,this paper is based on computer vision technology,adopts the non-contact measurement principle of visual target tracking,and uses a high-speed camera to build a circuit breaker moving contact tracking test platform.For the research on the motion characteristics of the moving contacts of circuit breakers,the twin network tracking algorithm tracks the moving contacts in real time and records their specific positions in each frame of images.To this end,this paper proposes a dual-template Siamese network target tracking algorithm fused with attention mechanism.The main work is as follows:(1)The high-speed image sequence dataset is expanded,and the screening and cropping of moving images are completed to meet the input requirements of the improved Siamese network and further improve the effectiveness of network training;(2)It is proposed to introduce a hybrid attention mechanism to help the network screen the most discriminative target features,use the spatial attention module to reduce the target irrelevant background and interference information,and the channel attention module to adjust the channel weights and enrich the deep semantic information of the feature map to further Improve the robustness and model generalization ability of the model to deal with the interference of target similarity;(3)The dynamic template branch is proposed,and the confidence update strategy is used to update the dynamic template adaptively,which avoids the tracking error caused by the deformation of the target appearance.Under the condition of retaining the complete information of the target in the initial frame,it plays an important role in the target template.The complementary and auxiliary functions enhance the robustness of the Siamese network model.The experimental results show that the algorithm in this paper has fast tracking speed and high accuracy,and can effectively deal with the problems of occlusion and blur caused by the vibration of the circuit breaker body and the rapid movement of the target,effectively improving the robustness of the tracking model and the tracking success rate,and realizing the circuit breaker.Robust visual object tracking in complex scenes. |
