Advanced Materials Research Vols. 712-715

Abstract: Camera calibration is the basis of vision-based 3D measurement. While many calibration methods have been proposed, the problem encountered in the practice of camera calibration is how to get accurate calibration parameters, which is seldom involved in references. This paper is focused on investigation of main factors influencing calibration accuracy, including manufacturing error of calibration rig, extracting error of control point and their combination. Based on the popular calibration method, simulation experiments are conducted at different error level, and the results show that the extracting error of control point has greater effect on calibration accuracy than manufacturing error of calibration rig. The manufacturing tolerance of calibration rig and extracting tolerance of control point is suggested to satisfy usual machine vision application.

Abstract: In order to get texture features of color image, we proposed a new algorithm which is adapted to extract the texture features for color images in this paper. Firstly, the proposed method adopt quaternion to color image processing that can represent the color image in a holistic manner and parallel processing the R, G and B components. Secondly, we can obtain the directional information and texture features by Quaternion Gabor Filter. The experimental results show that texture feature obtained by our method has good the discrimination power and classifing performance.

Abstract: The result of license plate recognition with a single feature is unsatisfactory. A multi-feature fusion method based on D-S evidence theory is proposed to improve results of mine loadometer license plate recognition. Firstly, three kinds of features including contour, projection and trellis-coded are extracted from the vehicle plate character image. Then the Basic Probability Assignment (BPA) is defined to get the credibility of recognition results by using the multi-class Support Vector Machine (SVM) with one-against-one method. Finally, D-S evidence theory is employed to integrate the credibility of evidences for making a final decision. The experimental results show that the multi-feature fusion method has higher recognition rate, fault tolerance and robustness.

Abstract: A motion object detection method is presented based on Davinci platform. This paper adopts color histogram algorithm to detect moving target, which is operated on TMS320DM6446. In continuous frames, probability distributions of both foreground pixel and background pixel are counted separately to build color histogram. Background probability can be computed based on gauss model. After background separation, we can use median filtering to suppress image noise, detect connected domain, converge foreground pixels to make up of moving object regions and mark them. The results show that the method presented has good accuracy and quick speed for realtime application.

Abstract: Fuzzy C-Means clustering(FCM) algorithm plays an important role in image segmentation, but it is sensitive to noise because of not taking into account the spatial information. Addressing this problem, this paper presents an improved suppressed FCM algorithm based on the pixels and the spatial neighborhood information of the image. The algorithm combines the two-dimentional histogram and suppressed FCM algorithm together. First, construct a two-dimentional histogram instead of one-dimentional histogram, which can better distinguish the distribution of the object and background for noisy images. Then determine the initial clustering based on two-dimensional histogram. Last, provide a new way to determine the suppressed factor and use the improved FCM algorithm to realize the image segmentation. Experimental results show that the improved algorithm is effective to improve the clustering speed, and can achieve better segmentation results.

Abstract: In order to solve the problem of highway intelligent video surveillance system for effective monitoring of vehicle operating conditions, a fast block background modeling method is proposed in the framework for intelligent video surveillance system. First using statistical histogram to build the background model of the video surveillance system, second using background subtraction method to locate the moving target area, at last using displacement of the minimum exterior rectangle centroid of the moving target between two frames to calculate moving target speed, without the aid calibration. Experimental results show that the proposed method exhibits its superiority in processing time, the time of building background model through 100 frames is 3.8s. The proposed method has good practical value used in intelligent video surveillance.

Abstract: This paper describes the use of image processing technology on the measurement of Arabidopsis leaf areas. Firstly, a black and white checkerboard was used to correct the image. This was done by finding the corner of the grid and performing affine transformation. The corners were also used to calculate the number of pixels of the checkerboard grid. Secondly, the plants were segmented with the values of the red, green, blue components of each pixel. Finally, the areas were calculated with the number of the pixels of the plants and the checkerboard grid in the image, as well as the real area of the board grid. The result has shown that this non-destructive method can reflect the growth rate by measuring the areas of the Arabidopsis leaves.

Abstract: To automate present Automatic Optical Inspection (AOI) systems, an intelligent method based on lightness analysis for solder joints detection is proposed in this paper. Firstly, the image taken from real printed circuit board (PCB) is preprocessed and segmented. Then, we compute the image similarity by calculating the Euclidean distance. If the similarity exceeds the threshold, we extract the lightness of solder joints by converting between RGB color mode and HLS color mode. Lastly, all the defects of solder joints are classified into several different types according to their lightness appearances. Experiment results show that the method deduced in this paper can detect accurately solder defects from different patches.

Abstract: In order to improve the rotation invariant property of the conventional template matching method, a novel template matching based on the polar coordinate was proposed. The origin of the polar coordinate was at the center of the template. And the matching result was central symmetry, which made the method have the translation and rotation invariant properties simultaneously. The recognition process was divided into two phased. In the first phase, gray information was used to complete matching calculation, and some candidate points were selected according to the matching result. In the second phase, edge strong matching of candidate points was completed. The candidate point, where the sum of gray matching value and edge strong matching value was minimal, was determined as the best matching point. Experiments results show that this method can meet the real-time requirement of the TV tracking system.

Abstract: About the key problems of the binocular vision system design which was used to video log examining, and according to the binocular parallax theory and triangle measuring theory, the paper researched the basic principle of camera imaging, proposed the camera parameters calibration method that could determine the best placed baseline distance of two cameras and the camera focal length together, inferred the relationship of the best baseline distance and the camera focal length by analyzing to many groups calibration results. The simulation experiment results show that this method is simple, practical, and has a high precision, provides a theoretical basis for design of tow cameras positions in the binocular vision system.