| With the rapid development of the national grid,the current manpowerbased operation and maintenance model is gradually unsustainable.Therefore,in recent years,a large number of robots and automation technologies have been researched and applied in the power grid.However,the inspect robot currently applied in power substation can only execute the inspect tasks,which means the operation tasks cannot be completed without staff on site.These operation tasks usually need to operate several power distribution switches in strict order.To further reduce labor costs in power grid operation and maintenance,this paper focuses on the visual positioning and recognition system required for the operation robot.In the indoor scene of the substation,there are two main types of distribution switches that need to be operated frequently by the operation robot: the square distribution switch and the circular earthing switch.In this paper,a complete binocular vision based positioning algorithm is proposed for these two types of power distribution switches respectively.According to the requirements of the power distribution room duty-keeping robot,a general framework for safety-oriented visual applications is also proposed.using a large amount of labeled data automatically collected by this framework,the accuracy and stability of the above two algorithms are verified.For the square distribution switch,this paper proposes a positioning algorithm based on edge line fitting.The algorithm fully combines the accuracy advantage of the machine learning algorithm on the classification task and the precision advantage of the traditional image algorithm in the positioning task,as a result,it can quickly and accurately find the correct square distribution switch in the picture and recognize it status.In the positioning process,the algorithm uses a special convolution kernel to process the contour and fit the feature points with a large number of contour points to obtain strong anti-interference performance.In the fitting process,the global accuracy of the Hough transform method and the local optimality of the least squares method are combined,so that the fitting process converges quickly and accurately to the global optimal solution.The stability of the algorithm verified by a large mount of tests.The positioning success rate reaches 100% in the test.The relative translation accuracy can reach 1.5 mm,and the relative rotation accuracy can reach 1.5 degree,which satisfy the requirements of the operation robot.For the circular earthing switch,this paper proposes a positioning algorithm based on ellipse fitting and least squares.The algorithm makes full use of the characteristics of the circular earthing switch,and disassembles the positioning process into two parts: the positioning of its circular contour and the determination of the rotation angle of its central cuboid.In order to obtain the spatial parameter equation of the precise circular contour,the algorithm first uses the RANSAC algorithm to solve the elliptic equation in the image,then uses the algebraic method to directly resolve the approximate parameter equation of the space circle,and then solves the optimal parameters by least squares method.The entire pipeline is designed to guarantee the anti-interference ability and stability.In order to determine the rotation angle of the central cuboid,this paper combines RANSAC algorithm and Hough transform algorithm,and fully considers the influence of light and shadow and projection distortion,and designs a high stable method to solve the central cuboid rotation angle.The stability of the algorithm can meet the requirements,which is verified by a large mount of tests.The positioning success rate reaches 84.5%,the relative translation accuracy can reach2 mm,and the relative rotation accuracy can reach 2 degrees,which satisfy the requirements of the operation robot.In this paper,based on the special requirements of the accuracy and stability of visual applications for power distribution operation robots,a general safety-oriented framework for visual applications is proposed.This framework accommodates all visual applications and provides a uniform interface.The framework verifies the results returned by each visual application by using a priori information and repeated results to ensure the accuracy of the visual application’s result.In addition,the framework has a complete internal error handling process that can handle most of the visual program errors internally without the intervention of other modules,thus improving system stability and simplifying system complexity.The framework also provides automatically labeled data collection,to quickly capture large amounts of data for training visual program algorithms.Based on the automatic labeled data collection capability of the framework,this paper proposes a standardized,automated and rapid visual application evaluation method,which relies on a large amount of data to obtain the stability and precision analysis of each visual applications. |
PDF Full Text Request