Research On Key Technology Of Steel-bars Separation Based On Computer Vision And Its Application

Although computer vision technology is assumed to have wide prospects of application, the scope of its current application in the industry is limited because of many unsolved theoretical and application problems. The main problems are arisen from the incongruity between its current techniques with respect to sequence image processing, object detection and the requirement of rapidity and accuracy in the online control industry. Therefore, it is a big challenge for computer vision researchers that the industry is asking for realtime, adaptive, anti-jamming and low-price vision systems.A theoretical and practical study of industrial machine vision is conducted in this dissertation. By addressing the authentic problems of online counting and separating steel bars in production processes, it primarily aims to investigate the issues concerning the rapidity and accuracy of object detection, multi-object tracking, and feedback control. This study solved a series of technical problems in relation to automatic control of online counting and separating. The main achievements of it include the advances in process technique integration, quasi-circular object recognition, object motion estimation, multi-object tracing, tolerant counting and vision feedback control. That makes it to have a significant contribution to the research and application of online vision detection and control.First, this study analyzed the existing steel bar counting techniques, evaluated the main problems existing in the variety of the current relevant models, and created an approach to machine vision-based object inspection, tracking, counting and separation. Based on the idea of information fusion, a specific mechanism of visual information processing integration by adopting a general theory of processing technology integration, was established aiming to deal with the difficulties and concerns resulting from the transfer of the visual information processing technology into the applications in the industry.Then, a fast iterative algorithm by making use of slide window overlapping was proposed on the basis of the studies of fast algorithms for pattern matching in object recognitions. In order to satisfy the needs of the practical applications of machine vision to object detection in the industry, an elaborative formula was developed in this study to implement a fast computation of real-time pattern matching in that the issues of multi-value pattern matching can be resolved. Meanwhile, a fast maximum-point detection algorithm was proposed for object location.To realize the function of size and direction independent detection, a quasi-circular object recognition algorithm was put forth in this study by taking advantage of the similarity among the polar contour coordinates. Furthermore, an annular zone edge convergence algorithm was proposed aiming to detect the quasi-circular object, and based on the idea of edge direction statistical inspection and the dual of circle center and circle perimeter, edge centralized algorithm was also come up with, which converts the passive statistic to active center converging. As a result, these algorithms realize the automatic edge separation and greatly improve the speed of computation for approximately circular shape object center enhancement.Aiming to solve the problems of multi-object tracking in the process of piled steel bars movement, the projected curve was used to estimate their global offset and to predict their new position in this study. Also, it was proposed that the accumulated projected curves can remove the prediction error caused by individual steel bar's sliding. Furthermore, in order to track the steel bars between the images, iterative position matching method was proposed. According to the defined dynamic observe confidence, a novel method called K time tolerant counting was put forward based on the repeated observations. By doing so, it is proven that the reliability of systemic counting is improved, and the negative effects caused by miss-recognition and omitted recognition are reduced.Based on computer vision servo via extracting and the position of the steel bars and splitting axe, a visional feedback separating control scheme was proposed that enables us to observe the states in complex steel-separation processes. Through analyzing the pattern of steel bars movement in the separating process and by putting forth a mechanism that simulates the steel bar moving by adjusting new position with restriction, this study created an emulation algorithm for steel separation.Based on the numerous data obtained from computer simulation, a strategy for controlling steel separation was used by means of fuzzy combination-based data mining techniques. The acquisition of the emulation and control strategy will provide the foundation of intelligent learning for steel separation in the real processes, make it possible for smooth steel separation control, and realize the vision-based state feedback control.This thesis discussed the techniques about experiment, offline debugging and online application. A great effort was made to improve the system structure, algorithms and strategies for the problems arising from the practical applications. Finally, the main technology ideas, research methodology, and innovative points emerging in this study were summarized, and several suggestions were made for future research.