Search

According to the current status for measuring water-level of the river, a river water-level remote monitoring system has been designed, which consists of a water-level sensor, a single chip microcomputer, a wireless module and user terminals. The designed contact type water-level sensor based on a simple principle with a high performance-cost ratio, has eliminated the shortcomings of the current sensors, which must be placed under the water with difficulty of water proof. The water level information collected can be real-time transmitted with signal of GPRS to the server terminal, users can easily transfer real-time data and history of data record from the server terminal. The experimental results show that the system can work stably, with a high accuracy, low cost, and easy to market.

This paper presents a moving object detection method with cast shadow suppression in the RGB color space. Firstly, the background model is constructed by the metrically trimmed mean and updated adaptively. Then, the information of brightness, texture, and chromaticity is fused to discriminate the shadow and the moving object. Our experimental results demonstrate the good performance of the proposed method.

A face tracking algorithm based on LBP (Local Binary Pattern) histograms and particle filtering is presented in this paper. The mean shift algorithm is adopted to locate the face region coarsely. Then the particles can be propagated efficiently with small random noise added. Since color histograms as face representation can be disturbed easily by the background with skin color. LBP histograms are used as the representation of faces. In order to alleviate the influence of light variance, the face template is updated under certain condition. The experimental results show that the proposed algorithm is superior to the CAMSHIFT algorithm and the particle filter based on color histograms.

With the extensive application of digital television, video compression technology has become one of the topics which the telecommunication industry concerns most. A video compression method based on redundant discrete wavelet transforms, Delaunay triangle and triangular affine transformation is presented. The feature points and potential motion areas (PMA) are extracted in the wavelet transform domain, and motion estimation is done in the time domain. Experiment results show that the algorithm can increase the PSNR value and reduce the video coding complexity. It has a certain practical value.

In this paper, a region of interest (ROI) MRI image coding scheme is proposed. This scheme use the Maxshift ROI image coding method supported in JPEG 2000 image coding standard. An automatic ROI detection method is also introduced. Experiments results demonstrate that the proposed method has good performance on MRI image compression.

For the requirement of monitoring cross road violation in intelligent traffic system, a method to recognize and track the peccant vehicle is presented. The static background is modeled by mixture Gaussian model, and the location of lane line is detected by Hough transformation, thus, the coordinated series can be obtained from the monitor image. The information of vehicles can be obtained by background-frame difference method, and according to the distance between the vehicle and line, the peccant vehicle can be detected. A mean-shift method is used to track the peccant vehicle, and the close range camera is used to snapshoot the license plate according to the center of tracking window. Actual road tests show that the work efficiency of the method is high, and the accuracy is up to 80%; so it has a certain practical value.

A method for moving multi-target recognition and tracking in stable scene is presented. Optical flow is used to extract the velocity of pixels, and targets are recognized by combining motion character points obtained by binary discrete wavelet transforms (BDWT). A discrete kalman filter is used to track targets in the follow-up frames; the center and scale of tracking window are updated by a Mexico wavelet kernel function mean shift method which is embedded into the discrete kalman filter framework to stabilize the trajectories of the targets for robust tracking during mutual occlusion. The method is tested on several frame sequences and shown to achieve robust and reliable frame-rate recognition and tracking.

Moving vehicle recognition and tracking is the key technology in the intelligent traffic monitoring system. For the shortcomings and deficiencies of the frame-subtraction method, a binary discrete wavelet transforms based moving object recognition algorithm is put forward, which directly detects moving vehicles in the binary discrete wavelet transforms domain. For the shortages of RGB or HSV color space based vehicle shadow segmentation algorithms, shadow segmentation algorithm based on YCbCr color space is proposed. First, the motion area which includes the vehicle and the shadow is selected by binary discrete wavelet transforms, and then the original data of the shadow according to the characteristics of the occurrence of shadow is chose, finally, the shape and location of the vehicle region is determined. An automatic particle filtering algorithm is used to track the vehicle after recognition and obtaining the center of the object. The actual road test shows that the algorithm can effectively remove the influence of pedestrians, cyclists in the complex environment, and can track the moving vehicle exactly. The algorithm with better robustness has a practical value in the field of intelligent traffic monitoring, and it is adopted by Tianjin Traffic Bureau.

A redundant wavelet transforms based stereo video objects segmentation algorithm is presented. First, the redundant wavelet transforms is used to extract the feature points of stereo video images, and then according to the feature points, the disparity estimation is done to form a disparity map. The stationary video objects are segmented from the stereo images by disparity map. For the moving video objects,a redundant wavelet transforms based moving object extraction algorithm is used to segment the moving objects from the redundant wavelet domain.Experimental results show that the algorithm can segment video objects from stereo video images,including stationary objects and moving objects with good accuracy and robustness.

Vehicle license plate is a special region of vehicle image, which has many features and can not be located by only one of these characteristics. Based on multi- characteristics a novel real-time license plate detection method is presented, which consists of two main steps: license plate area's rough detection and license plate segmentation & verification. The former is stimulated by the fact that the front characters' color and the background's color are didymous and their run- lengths are alternant and restricted, while the latter is based on edge and size of license plate region. Experimental results show that this approach not only has a high locating rate, but also can achieve the realtime processing since it costs 20 ms only.

The segmentation of video object is an important technique of multimedia processing. The biggest characteristic of MPEG-4 which is the multimedia compressing standard is that it puts forward the video object based compressing. Currently, there is no method which can segment the moving object effectively. This paper puts forward a new method to obtain the motion area based on redundant wavelet transforms and then to segment the moving video object. Experimental results show that the method can obtain the moving object effectively and with strong robustness.

Locality preserving projections (LPP) is a linear method that optimally preserves the local structure of the data set. However, because of the continuing existence of transformation difference, LPP subspace is failed to detect the important nonlinear variations of the face manifold. In order to improve the recognition performance, we propose to use radial basis function network (RBFN) to classify the features in the LPP subspace. The multi-quadrics function is taken as the activation function of the RBFN and the hidden layer of RBFN is trained via the gradient descent algorithm in our approach. The experimental results show that the LPP+RBFN method has achieved a better rate than Laplacianfaces, Fisherfaces and Eigenfaces.

Due to the large requirement for memory and the high complexity of computation, JPEG2000 cannot be used in many conditions especially in the memory constraint equipment. The line-based wavelet transform was proposed and accepted because lower memory is required without affecting the result of wavelet transform. In this paper, the improved lifting scheme is introduced to perform wavelet transform to replace Mallat method that is used in the original line-based wavelet transform. The three-adder unit is adopted to realize lifting scheme. The corresponding context-based arithmetic coding is designed here so that the proposed algorithm is more suitable for equipment.