Research And Application On Object Location Of Digital Image Processing Based On Multi-Information Fusion

In the relative research of metallurgy process parameters measurement such as vision-based rebar counting system, molten steel level measuring system and billet surface temperature measuring system, the high temperature, dust, vibration, electromagnetic interference, complex background interference and space restriction may lead to serious noise, hash and information missing in the digital image data of vision-based measuring system. In such a case, traditional object extraction methods can not perform reliably for short of restriction abilities under severe interferences.An object extraction method based on multi-information fusion is proposed here to solve the object location problems caused by the severe interferences which can enhance the reliability and stability of vision-based metallurgy process parameter measurement system. Discussed with the four main forms of digital image processing-single-source image, video stream, data and mechanism model, multi-sensors, this method has been proved to have the ability to ensure the object location reliability by feature decomposition and reconstruction.The main innovations in this paper are as follows:1. Object extraction from single-source image based on multi-information fusion methodAn object extraction method is proposed here especially for unstable-object positioning by pixel-level fusion of the template matching data and the multi-threshold segmentation data. By making fusion of the gray-scale distribution, shape and gray-scale level, the objects in the single-source image are reliably located even when they are unstable in brightness and shape. Respectively, in order to reduce noise, an improved median filter algorithm called slider-choiced filter is used to enhance image quality. Having been used in vision-based on-line rebar counting and separating system, this method is proved to have the ability to overcome the interferences of rebars’ oxide and inter-cover. The position rate is higher than 99% in the case of brightness attenuation rate is <50% and section inter-cover rate is <70%.2. Multi fast-moving similar objects tracking in video streams using multi-information fusion methodAn object extraction method based on cumulative projection and extended Kalman filter prediction (EKF) fusion is put forward for multi fast-moving similar objects tracking in video streams. First of all, an inter-frame offset prediction is made by cumulative projection method. Then, each independent rebar section object is added to the tracking chain as an EKF node to overcome the miss-matching problem caused by large inter-frame displacement. Finally, the reliable prediction of target position in the video stream is realized. Having been used in bar counting and seperatating system of bar plant, this method is proved to be reliable and the track success rate is >98% when the inter-frame displacement is less than 1.5 times of the diameter of the rebar section.3. Faint object Location by mechanism model and image data fusion methodA method for faint object location by fusing mechanism model criterion and image data is proposed to enhance the reliability of faint object extraction in the complex noise and interferences of the industrial field. By analyzing and estimating the state of the object from the mechanism model, the criterion is obtained and used to eliminate the fake objects formed by the noise and interferences. This method is been used in the interface location between the slag layer and the molten steel layer in molten steel level measurement. It can overcome the interferences of both noises and slag, and the interface location error is less than 5mm.4. Feature extraction based on multi-sensors data fusion technique and automatic data registrationAn automatic data registration method is proposed for not-pre-located data of different dimension sensors based on trend distribution analysis. Been used in the surface temperature measurement of continuous casting billet, it can give a reliable way for data registration of CCD camera and infrared radiation thermometer.In a word, the multi-information fusion method can perform reliably in harsh environment of metallurgical industry by improving information content and noise immunity. Proved by field applications, multi-information fusion method is suitable and will have a bright application prospects in metallurgical process parameter measurement.