Visual Detection Method For 3D Shape Of Soft Manipulator

Due to the characteristics of continuous deformation of any part and flexible structure,soft manipulator is suitable for small and complex environment,and has great application potential in medical,disaster relief,aerospace and other fields.In order to realize the closed-loop control of the soft manipulator,the shape information of the manipulator needs to be obtained.Based on stereo vision,this paper has carried out the research on the three-dimensional shape detection method without marking points for the pneumatic soft manipulator.The main research contents of this paper are as follows:Firstly,a three-dimensional shape detection algorithm based on Self-organizing Map(SOM)is designed and implemented.This paper uses the MATLAB calibration toolbox to complete the calibration research of the binocular camera.After that,this paper conducts experiments with hose as the shape detection object,compares the SOM algorithm with the commonly used clustering algorithm K-means,and analyzes the advantages and disadvantages of the two algorithms from many aspects.After determining the use of the SOM algorithm,three error indicators are used to evaluate the clustering effect of the SOM algorithm.Through multiple rounds of experiments,the relationship between the SOM algorithm parameters and the shape clustering error is analyzed in detail,and based on this research,SOM algorithm parameters are designed for the hose and data clustering is completed.Finally,the three-dimensional shape of the experimental object in the camera coordinate system is reconstructed by the triangulation method,which verifies the effectiveness of the algorithm.The algorithm does not need to mark the target in advance,nor does it need to use the three-dimensional point cloud data of the target.It only needs to ensure that the target does not overlap and occlude in the camera's perspective.Secondly,in order to estimate the deviation between the detection result of the algorithm and the real shape,a mark-based three-dimensional shape detection algorithm is designed and implemented.This paper keeps the shape of the hose unchanged,and marks it at equal intervals along the direction of the hose.After that,an image processing algorithm was designed to obtain the marked points on the hose,and the sorting of the marked points and three-dimensional reconstruction were completed.In order to facilitate the calculation of the error,this paper performs a polynomial fitting on the detected three-dimensional point set,takes the result as the standard value,and compares it with the detection result using the SOM algorithm.In addition,the average deviation is defined and the error is quantified.It can be seen from the results that the proposed detection algorithm based on SOM can achieve higher accuracy.Finally,the soft manipulator motion platform and the real-time detection algorithm of three-dimensional shape are designed and implemented.Through analysis and research on related components,a soft manipulator movement platform is designed and built.In addition,based on the previously proposed SOM-based shape detection algorithm,considering real-time requirements,the algorithm is optimized from three aspects,and the changes in the detection accuracy and training speed of the algorithm before and after optimization are analyzed.Finally,the built motion platform is used for experiments,and the real-time detection of the three-dimensional shape of the soft manipulator is completed,which improved the speed of the algorithm while meeting the accuracy of the shape detection.