Research And Implementation Of Measurement Technology For Specific Components Based On Machine Vision

Machine vision as the "eyes" in the modern intelligent industrial, via using image processing methods on image which captured by image acquisition device to acquire the object feature. Then through object detection and measurement method assist intelligent equipment complete relevant operation. There are more important research significance and application values in industry for specific components measurement, due to specific work-pieces have irregular structure, extraordinary pose and complexity related influencing factors. It could not use the traditional pose measurement methods in machine vision system. This paper proposed a method using dual-thresholds for image segmentation and a method using planar features in monocular machine vision to pose measurement methods to complete the measurement of specific work-pieces.In order to deal with the segment problem of dynamic image which have indistinctive ups and downs feature in histogram, a method of dual-threshold dynamic image segmentation base on DTBNN (Decision Tree Based Neutral Network) algorithm is proposed. Firstly, according to the correspondence between decision tree and neural network this paper build a stable and efficient training neural network; Then, the method work out the mean of gray value, the maximum deviation and the threshold mapping function as sample data are used to train the neural network; Finally, this method using the trained neural network to get threshold mapping function by testing images, and complete dual-threshold image segment by using the result of the upper and lower threshold values which are calculate from previous step. The simulation results show that this method does not depend on the histogram feature and it could accurately get upper and lower segmentation threshold. Compared with Otsu dual-threshold method and maximum entropy dual-threshold method, the proposed method in this paper could achieve better dynamic image dual-threshold segmentation.In order to realize two coordinate systems in space of pose and position measurement for monocular vision, this paper approach a pose measurement method based on planar property of two-dimensional machine vision image. Firstly, this paper analyze the pose measurement method based on coplanar points properties and coplanar lines properties; Then, it convert the pose relationship problem into the problem of relationship between each directional vector of planar in two coordinate systems space, from above operation could obtain the resolution component of rotation matrix; Finally, Through solving the rotation matrix components in each rotate angle, this paper could obtain the relative pose results in two planar. This paper using captured original image in practical project as input data; After that it via the analysis of geometry proving its feasibility; At last, this paper compared with the measurement method of based on coplanar points properties pose measurement method and based on coplanar lines properties pose measurement method in same projection. This method proposed above has good measurement performance, less error rate in practical engineering, and satisfies accuracy requirements of projection in real-time processing.This paper applied methods mentioned above to the measurement process for special components in the actual engineering, which are dual-threshold dynamic image segmentation base on DTBNN and using monocular machine vision based on planar property to pose measurement. Through actual projection testing show that it as good applicability in application, could get an efficient measurement results, and achieve engineering requirements of measurement accuracy.