Research On A Novel Method Of Measurement Based On Computer Vision And Its Application

The final target of computer vision research is to make the computer capable of adapting itself to the ambient condition by observing and understanding the external environment with its visual sense. At present, a universal computer vision system, which can identify the objectives in the random background and can independently accomplish various works, could not be developed. But it is possible to devise a special computer vision system that can automatically complete a specific work. In this paper, several measurement approaches based on the computer vision were studied and put forward in order to measure the steel rods online in the steel and iron industry. And then a method for measuring the steel rods online, which integrated interdisciplinary knowledge in computer vision, probability and statistics, pattern classification, and information fusion, has been developed. At last, based on the developed method, a computer vision system to measure the steel rods online automatically has been devised. The main research achievement is showed as follows:1. Concept of virtual multi-sensor was advanced, and a method of pattern classification via the information fusion using weighted coefficient Dempster-Shafer (D-S) evidence theory was put forward and applied to the computer vision measurement. Diverse feature pictures could be obtained by means of processing and transferring the real-time image with different methods. By scanning feature pictures on the same position with templates at the same time, degrees of basic belief, each of which implies that there exists a measured objective in the detected region in the scanned picture, could be obtained. Then fusing these degrees of basic belief using weighted coefficient D-S evidence theory, a judgment whether the measured objective exists at the same position in the original real-time image could be given. After scanning all position in pictures, the total numbers of the measured objective in the real time image could be obtained.2. Information fusion algorithm using the self adapting weighed coefficient D-S combination rule, an edges detection method and a region segmentation method that the initial value of the parameters and threshold value are self-adaptively adjusted respectively all according to the statistical characteristics, have been proposed after studying the characters of the online disturbances. Based on the all of above, the self-adaptability of the multi-sensor information fusion and the anti-interference ability of the process to acquire the information have been all improved, and the degrees of the basic belief have been determined by the historical data and the statistical characteristics from the real time image. The technique of the multi-sensor management is applied to the measurement system based on the computer vision so that the computational complexity has been reduced and the measurement efficiency has been increased,3. Measurement method based on the virtual multi-sensor and support vectormachine (SVM) were brought forward. The detected region could be classified using the SVM and the information from the virtual multi-sensor so as to determine whether the measured objective exists in this region. By this means, the dependence of the measurement system on the prior knowledge about the specific engineering has been abated, so the practicability of the proposed method has been further increased.4. Automatic measurement system for detecting and counting the steel rods online based on computer vision was developed. After analyzing the disturbances on the images caused by rays of light from external environment, emission of radiation, a mechanical system that could automatically acquire the real time images from the complex dynamic environment in the industrial fields has been devised, so the stable images could be acquired. The developed measurement system has been applied in the steel industry to detect and count the steel rods online. The result of this application shows that the measurement accuracy arrived at 99.6%.