Advanced Materials Research Vols. 694-697

Abstract: An independent hardware system based on monocular vision has been built to measure the posture of moving objects suspended. The infrared light source and infrared filter were used to reduce image noise and interference. The six feature points distributing on the different surfaces of the object suspended were determined by PnP problem theory. According to the perspective projection constraints of the 3D coordinates and image coordinates, three posture angles were calculated as the object suspended moved. A verification test has been carried out at the three-axis-turntable to verify the static and dynamic characteristics of the measure system. The results show that the monocular vision system constructed has a good stability of higher measurement accuracy and certain dynamic instantaneity, and may be applied to measure the dynamic posture of the objects suspended.

Abstract: Skin colour is an important attribute that can be used to detect human presence in an image. In this paper, a new method is introduced to detect skin pixels in an image based on statistical decision theory. The proposed method uses a parametric model to produce a pixel seed that represents a potential skin pixel. Then the validity of the pixel is confirmed using a moving window and the nonparametric sign test. Experiments show that the proposed method can produce very accurate skin detection results.

Abstract: Self-calibration of stereo rig is essential to many computer vision applications. In this paper, a new self-calibration method is proposed for a binocular stereo rig undergoing a single motion with varying intrinsic and extrinsic parameters. Firstly, we build up a stereo rig model based on the basic platform to describe the transformation of the stereo rig during the motion. Secondly, the characteristics of singular values of the essential matrix are used to estimate the intrinsic parameters of camera. Finally, analyzing the transformation relation between different views, the relative position of cameras and motion of the stereo rig are estimated. Experimental results for both synthetic data and real images are provided to show the performance of the proposed method.

Abstract: An effective approach of quality inspection for micro hole is proposed by studying on the geometric feature of micro hole edge. The edge of micro hole is detected exactly with the digital image processing techniques, such as, the median filter, the threshold value transformation, and so on. The least square circle fit arithmetic and a new roundness measurement arithmetic are synthetically used in the processing of geometric feature recognition of micro hole. The practical results prove that the method with exact conclusion and high precision can satisfy the quality inspection requirement of micro hole.

Abstract: Through the extraction of face image Gabor feature, combined with Adaboost for face recognition. According to the characteristics of high dimension Gabor, redundancy is large, the introduction of Adaboost algorithm for feature selection to reduce the dimensions of feature vector, for a large number of Gabor feature selection. At the same time ,using a single positive sample set and several negative sample sets for training method to construct a strong classifier cascade classifier. Testing results in the Yale library proves the validity of the method.

Abstract: Capturing screen technology can be applied making interactive software to serve skill training in the enterprises or engineering teaching in universities. This paper discusses the technology of capturing screen and its application fields. Some problems which usually may be met during applying capturing screen technology and producing video file are discussed. And related methods are given to solve these questions. The steps of making interactive software are proposed. Some good suggestions are given about how to construct software, how to realize interactive function, and how to process video file. These methods of solving problem and suggestions summarized from our work experience will be helpful to improve effect of skill training or engineering teaching.

Abstract: Most of the methods, which being used to the normal light, are easy to make a mistake in the high light condition. An improved method is proposed in this paper to solve these problems. It includes a series of pretreatment for road image, like white balance, contrast enhancement, edge enhancement. Then the method makes an image segmentation to increase the efficiency of the identification. Use an improved Hough transformation to recognize the parameters of lane line. Finally, establish a trapezoidal interested region to achieve a real-time dynamic extraction of lane lines parameters from the continuous image. The results of identification show that the improved method for the high light condition makes a better work and its more accurate and efficient to acquire the parameters.

Abstract: It is a key step to extract moving objects from background for computer vision applications. GMM based methods are the most commonly used technique for background subtraction in video sequence. However, how to establish efficient and precision background model with fast convergence rate is a Research-Worthy problem. In this paper, an effective scheme is proposed to accelerate the convergence rate of Adaptive-K Gaussian Mixture Model (AKGMM). The AKGMM algorithm alters the dimension of the parameter space at each pixel based on the changing frequency of pixel value. The number of GMM reflects the complexity of pattern at the pixel. An improved learning method is proposed for Gaussian Mixture Model. An adaptive learning rate is calculated for each Gaussian at every frame for speeding up the convergence without compromising model stability. Experimental results demonstrated that the proposed method gets a faster convergence while maintaining good robustness against complex environment compared to a conventional method.

Abstract: Robot location is a key technology of quadruped robot with Hand-fused-Foot. The location method based on binocular vision system is studied for quadruped robot with Hand-fused-Foot. After obtaining image by a single camera, the object is segmented by using characteristic extraction method based on color characteristic. Then image processing such as filtering (de-noising) and opening is performed. The object is identified and its centroid coordinate in image is obtained. Location of robot based on environment reference--object coordinate is achieved. Experiments show the effectiveness and the accuracy (within 4cm) of the method.

Abstract: In order to overcome the accumulated error in traditional localization methods for intelligent vehicle such as dead reckoning and visual odometry, a simultaneous localization and mapping(SLAM) algorithm based on stereo vision was presented in the paper. Firstly, the interrelated elements in the localization method were defined, and the probability model for intelligent vehicle localization was proposed, then the motion and observation model were established, and the detailed implementation of the introduced localization algorithm was given. Finally, an experiment was designed to confirm the effectiveness of the proposed method. Experimental results indicate that the algorithm can realize three-dimensional motion estimation of intelligent vehicle and can improve the positioning precision effectively.