Detection Of Electric Power Tower's Nest Based On Binocular Vision

Since the reform and opening up,China's economy has been exponential rapid growth.At the same time,there is a huge increase in the demand for electricity in all sectors of society.Therefore,in order to protect the long-term stability of the industrial and residential power supply,the regular inspection of the grid is essential.In view of the problem of missed inspection in the process of power line patrol.This thesis mainly aims at the research of the important technology link in the process of image stitching and image recognition,and finally realizes the detection of the Tower Bird's nest in the process of power line patrol.Firstly,this thesis introduces the development status,characteristics and application fields of the relevant theory,and introduces the imaging model of binocular vision and the transformation between coordinate system in detail.The binocular imaging system based on vertical parallel optical axis imaging model is designed and explains the basic principle of image stitching.Secondly,some related techniques applied in image preprocessing are analyzed experimentally,the image of gray equalization is obtained.The Gaussian filter,the mean filter and the bilateral filter are used to contrast the experiments in the process of processing Gaussian noise.Because the bilateral filter has the good edge-preserving effect,it wins in the contrast experiment.The Sobel algorithm which combines with non-maximum suppression algorithm is introduced to enhance the effect.Some commonly used image morphological processing operations such as open and close operation were experimentally validated.Then the ORB algorithm and the SURF algorithm are compared in the feature detection of power towers in the image.Using the BF(brute force matching)algorithm to match the feature and the optimization algorithm to optimize the detected feature points to remove a large number of error matching.Finally,the image of panoramic stitching is obtained by affine transformation and “Fade into the fading out” image fusion technique.Then,the Otsu threshold segmentation algorithm is used to separate the towers in a simple background.By analyzing the maximum variance value,the algorithm is not suitable for the segmentation of complex background.At the end of the thesis,using the first and second Cascade classifiers based on AdaBoost by training positive and negative samples to detect the tower and the bird's nest in the image then achieves the experimental purpose finally.Finally,the thesis summarizes the work and points out the research work that needs to be carried out further.