Research And Application Of Visual Localization Based On Edge Feature Extraction Algorithm

Digital image edge feature extraction technology is widely used in target detection,intelligent driving of vehicles,medical condition diagnosis,face recognition,industrial vision localization,etc.Therefore,image edge feature extraction technology has good application prospects and research significance.In this paper,we take image edge feature extraction as the research object,aiming to improve the localization accuracy and single-pixel edge response of detected edges,and combine it with its application to vision localization in industrial fields.First,the importance of image pre-processing algorithms is comprehensively analyzed by taking products in actual industrial vision projects as examples.Suitable image pre-processing algorithms can enhance image features of interest,suppress unnecessary noise,and improve the accuracy and robustness of edge feature extraction algorithms.Then,a modified Sobel direction operator edge detection algorithm based on multi-directional fusion is proposed on the basis of in-depth analysis of traditional image edge feature extraction algorithms,which extracts single-pixel edge features by changing the weights of the Sobel diagonal direction gradient operator and selecting appropriate gradient thresholds,and finally fuses the edge features in diagonal,horizontal and vertical gradient directions.The edge features extracted by this algorithm not only have good localization accuracy and continuity,but also have good single-pixel edge response characteristics and the flexibility of arbitrary combination of gradients.Then,a straight line fitting algorithm based on contour edges is proposed.The algorithm first uses the least squares method based on the circular fitting strategy to pre-fit the edges of straight line segments,performs straight line clustering by angle 1 degree as an interval based on the straight line edges obtained from the pre-fitting,and reorganizes and fits the clustered set of straight lines twice.The algorithm is relatively close to the traditional straight line fitting algorithm in the complex image background to fit the results closer to the actual product straight line edges,with good anti-interference and applicability.Finally,this paper illustrates the general process of vision system software and hardware scheme by taking the vision positioning system of an electronic device electric testing and discharging equipment as an example,and applying the algorithm of this paper to detect and process the target product’s image for real-time positioning.The production of this equipment for ten consecutive days is counted,and the target product positioning accuracy is99.27%～99.74%,which meets the actual production requirements and fully verifies the effectiveness of the algorithm.