The accuracy of moving object detection based on traditional Gaussian mixture model is not high. In order to solve this problem, an improved method for moving object detection was proposed, which could improve the accuracy of moving object detection by Gaussian mixture model with spatial local correlation. Firstly, the Gaussian mixture model was built for each pixel in image, and the algorithm of the updating the variance coefficients based on the number of model matching was used to solve the problem that the variance converged slowly in traditional method. Then, the energy function of the Markov random field was redefined and combined with the spatial local correlation, and an adaptive threshold was obtained for moving object detection. By using the public available test data set from IBM Research, the experiments were carried out. Test results illustrate that the proposed method can adapt to the dynamic scenes much better, and obtain more accurate results of moving object detection.

An active visual system for intelligent tracking of moving object based on Digital Signal Processing (DSP) was designed with Expectation Maximization (EM) algorithm and a novel zonal logical strategy for camera pan/tilt control was presented. The dada set of the moving object gray feature was treated as Gaussian mixture model which included gray pixel position，and the expectation was employed as objective function. The localization and shape size of object in every frame which could be achieved during the estimation process with the EM algorithm were regarded as auto-control signal of the camera. The experimental results by using our system with several real image data demonstrate that the proposed scheme is not only practical and easy to perform the moving object intelligent tracking, but also robust for tracking. The system test shows that the processing for 320 pixels ×240 pixels image achieves 20 frame/s.

An improved tracking algorithm is presented in order to solve the problem that traditional tracking algorithm based on Mean Shift often falls into local extrema and causes inaccuracy or even failure when tracking a camouflaged target. Firstly, joint histogram is used to enhance the description of camouflaged target’s character. Secondly, target description model is modified according to the discrimination between the target and local background during the tracking period. Finally, the weights of each part of the joint histogram are adjusted adaptively to ensure more accuracy and robustness to background motions. The updated algorithm is tested with image sequences where the colors for both the target and the background are similar and results show that it can achieve effective and exact tracking.

Target tracking based on the space transformation model can usually be solved by dealing with the nonlinear least-square optimization problem. The geometric optimization algorithm based on the vector space has more limitation for high nonlinearities of the projective transformation. The mapping between a Lie group and its Lie algebra can make us to utilize the specific properties of the target tracking to propose a second-order minimization tracking method for the projective-based geometric warps within the intrinsic geometric optimization framework. In this approach, the Hessian matrix needs not to be calculated, so the computation complexity is reduced. Comparative experiments with the algorithm based on the vector space and the Gauss-Newton algorithm based on the Lie algebra parameterization validate the feasibility and high effectiveness of our method.

Aiming at the detection and tracking of IR point targets in complicated sea-sky background, an algorithm for detecting and tracking targets based on the Triple Pipeline was studied. Making use of the first and the second pipelines, this algorithm could achieve effective filtering through quadratic correlation of binary IR images. The triple pipeline based on “OR” operation was used to search targets and mark their positions .The method discussed in the paper can detect and track targets reliably. In the practice of engineering, we used real IR images that were screened by opto-electronic tracker to validate the reliability of the algorithm. Experiment shows that practical results are accordance with emulational results, and the algorithm is feasible.

The fitting model of multi-layer thin films in ellipsometry analysis was established. An unknown film was investigated and the physical structure model of substrate/film1/EMA/film2/srough was applied. Based on spectroscopic ellipsometry after determining the optical constant of the substrate, the modeling method of increased layers from single to 3-layer was provided. The results indicate that the Cauchy dispersion model is available for optical transparent or weak-absorption thin films. The structure of the unknown film was fitted with a model of G (1.52)/2.031 2 (203.0 nm) 1.463 6 (170.1 nm) 2.079 1 (170.4 nm) /A by ellipsometry analysis method, and the result was confirmed by the transmission spectrum from a spectrophotometer. The analysis to optical constant and film thickness also show that the ellipsometry is effective for multi-layer thin films analysis.

Infrared smoke screen is an important countermeasure method against infrared imaging tracker. The degree of infrared smoke screen shielding on ship target directly affects the tracker’s performance. In this paper, the box counting dimension is used to evaluate the degree. The key problems involved in the computation of Differential Box Counting (DBC) are discussed, which are the range of linear scales and the precise number of boxes, etc, and then the precise computing method for box counting dimension is presented. By the proposed box dimension, the fractal dimension value of the jamming region of ship target image is gotten, which can indicate the degree of infrared smoke screen shielding on ship target and achieve the evaluation of shielding effect.

A feasible Constant False Alarm Rate (CFAR) detector based on Anderson-Darling (AD) test for multiple interfering targets and clutter edge scenarios is proposed and referred as AD-CFAR, which chooses reference cell length and detection strategy adaptively according to characteristic of input clutter serials. AD-CFAR exploits K-sample Anderson-Darling hypothesis test to censor clutter blocks needed for clutter power estimation, and then Anderson-Darling test is employed for distribution test of the resultant homogenous clutter to select the proper detection algorithm from the strategy library. Theoretical analysis indicates that the proposed AD-CFAR maintains similar performance in homogeneous clutter to CA-CFAR and achieves excellent detection and false alarm performance in multiple interfering targets or clutter edge situation even when both are present.

For single landmine detection with the passive infrared imaging technology of the landmine detection, the infrared images were processed according to the 1-D signal detection theory and the algorithms for modern signal processing. The signal models of landmine and background image were given by analyzing the theory of landmine detection on the infrared image and combining the characteristic of the background infrared image. Then, combined with the characteristics of the models’ unknown parameters, we estimated the unknown parameters through the algorithms of modern signal processing. When the landmine and the background image signal were determined, the generalized correlator was used to detect the landmine. Testing results with the single infrared image burying the target show that the method can mark the landmine clearly.

In order to investigate the control laws of the infrared radiation characteristics of the aircraft skin within 8 ~ 14 μ m-band, the physical and mathematical models for aerodynamic heating and the internal heat transfer were put forward. The infinite volume method was used for aircraft skin temperature distribution and reverse Monte Carlo method was used to calculate the infrared radiation characteristics emitted by aircraft skin. By calculating the infrared radiation intensity of aircraft-skin, the contribution of infrared radiation of aircraft components in a horizontal azimuth was analyzed. Finally, the control laws of the infrared radiation characteristics of the different aircraft components on skin temperature and emissivity were obtained.

An effective method for real-time pedestrian detection applied to far-infrared images is presented, which makes use of the characteristics of pedestrian regions in far-infrared images and is based on a hybrid classification features algorithm. First, two levels of statistical adaptive oriented projection methods based on the high brightness property of the pedestrian pixels are used to locate the Regions of Interest (ROI) and eliminate the “shadow” phenomenon caused by one level oriented projection method. Then the method combines the pedestrian’s shape-dependent and shape-independent features (including shape’s morphological feature, inertia-based feature and histograms of oriented gradients (HOG) feature) to describe the ROI in the round. Finally, Support Vector Machine (SVM) is applied to classify and detect the pedestrian region. Experimental results of several far-infrared image sequences show that the proposed method achieves highly accurate pedestrian detection by combining hybrid classification features, and can be employed in real-time applications.

The paper reports homemade Yb3+ doped double-clad photonic crystal fiber for high power fiber laser, and gives the preparation process, pivotal technology and critical technology of a multi-component glass Yb3+ doping, from which the core of the double clad photonic crystal fiber is made. The fabricated Yb3+ doped multi-component glass and double clad photonic crystal fiber are tested by a series of physical properties, thermal and spectral experiments. The result of calculation of multi-component doped glass shows appropriate quality (Yb3+-(2mol%)-SiO2 (70mol%)-MgOCaO-BaO-Li2O-Al2O3(2mol%)), and good double clad structure of a photonic crystal fiber and optical properties to provide material basis for China's self-developed high power photonic crystal fiber laser devices.

To improve the measurement accuracy, the main measurement errors of the 193-nm Phase-shifting Point Diffraction Interferometer (PS/PDI) are discussed here. The elementary configuration and measuring principle of the 193-nm phase-shifting point diffraction interferometer are introduced firstly. Then different kinds of measurement errors and their causes are summed up, including reference wave front error, geometrical separation induced error, phase-shifting error, grating error, and errors caused by CCD, laser source and fluctuating surroundings and so on. The magnitude and shapes of these measurement errors, together with the connections between these errors and the configuration parameters of the interferometer, are all obtained through detailed analysis and simulation. For the sake of avoiding or restraining these errors, some methods are put forward accordingly.

In order to solve some problems of static parameter detecting & measuring system for gun tube, we invented the engine common flat that adapted to different gun tube with measuring accuracy above 0.05% of axes orientation and above 0.1° of circle orientation. We designed the hardware common flat controlled by double MCU and the software common flat under the guidance of modular design. Three common flats connected with the flaw-detecting module which could detect all parameters of flaw to make the measuring accuracy of flaw’s length and depth above 0.1 mm. Three common flats connected with the measuring module of bore size to make the measuring accuracy of all parameter of bore size above 0.000 2 mm. The common flats also can connect with other measuring module to get other parameters of the gun tube.

In order to improve the accuracy of the whole flexible measurement system, a compensation method based on robot was put forward. Constraint equations were constructed by using distance error model, and the actual geometric parameters of the robot were solved, and then these parameters were used in the modified kinematics model. Car body points were measured by using the modified kinematics model and compared with the distance data of laser tracker. Experiments prove that, the repeatability error of the measurement system is stable and less than 0.1 mm before and after compensation, but the absolute error of the measurement system declines from 0.6 mm to 0.3 mm and the accuracy is improved greatly.

To obtain real-time terrain information in front of the cross-country intelligent vehicle and effectively detect the possible obstacles, an environment detection method on the basis of stereo vision sensor was presented. Firstly, local image enhancement technology was proposed to weaken the influence of varying illumination and guarantee that the corner features could be extracted with uniform distribution. Secondly, edge features were extracted to avoid miss-detection of obstacles. Thirdly, stereo matching flow based on multi-feature extraction was presented. Fundamental matrix was calculated based on Random Sample Consensus (RANSAC) method. Corner-based matching scheme and edge-based matching scheme were implemented independently to avoid disturbance. Continuity constraint was studied to eliminate pseudo matching. Finally, different cross-country scenes with varying illuminations were collected to test the robust performance of the method.

In view of the faults of conventional image fusion methods, a novel remote sensing image fusion algorithm with high spatial quality based on IHS transform and the decimated wavelet transform is proposed. The merged high frequency coefficients come from the high frequency coefficients of the histogram matched panchromatic image with the proposed fusion method, while the fused low frequency coefficients are obtained according to the fusion rule suggested in this paper. The merged image is acquired with the inverse wavelet transform and the inverse IHS transform. Experimental results show that the proposed algorithm provides a better tradeoff between improving spatial resolution and preserving spectral quality. The computational complexity of the proposed method is close to that of the conventional fusion approach based on decimated wavelet transform and much lower than that of the fusion method based on undecimated wavelet transform.

A multiplicative iterative blind image deconvolution algorithm was proposed, in which both estimated image and Point Spread Function (PSF) were assumed to follow Markov Random Field (MRF) models. In addition to natural preservation of non-negative constraints, the algorithm introduced a set of novel antinoise difference operators, which first carried out weighted summation and then difference. Alternating Minimization (AM) method was adopted to linearly decompose the blind image deconvolution that belonged to a class of nonlinear inverse problems into alternate image restoration step and PSF estimation step. During the alternate iteration, an acceleration method called as vector extrapolation was applied to the alternate iteration steps owing to slow convergence of the algorithm. Numerical experiment results show the proposed algorithm can rapidly realize the restoration of the degraded High Resolution (HR) image from China & Brazil earth resource satellite-02B (CBERS-02B) meanwhile effectively preserves detailed edges and inhibits noise.

Aiming at the request of education academies and mid & small-scale corporations, conventional UV optical source systems in stereolithography apparatus were replaced with Ultraviolet Light Emitting Diode (UV LED) in the paper, and a LED optical system based on a single UV LED was designed. In the system, the beam from the single LED was focused through a multi-lens optical focusing system, and the LED chip was controlled with close-loop control strategies in order to keep steady operation. Experimental results show that the optical system meets the design request and has the advantages of simple structure, long stabilization and maneuverability. Compared with the other light sources in the stereolithography apparatus, the outstanding characteristics of the system are lower cost and higher performance on the condition of meeting speed and accuracy.

Fiber Optic Gyro (FOG) has been widely applied in stabilization loop of electro-optical tracking and pointing system, but the time delay unit caused by signal processing and transfer delay in FOG constraints the bandwidth of stabilization loop. On the basis of comparisons among limited memory filter, Kalman filter and least square method, least square method is adopted to forecast the measurement output of FOG, which compensates its time delay unit. In light of the measurement requirement of FOG and characteristics of time delay, factors affecting compensation accuracy are analyzed, and the degree of polynomial fitting of different least square method is tested. The result shows that the time delay unit of FOG can be well compensated by using the least square method of five-point and four-degree extrapolating two points in the frequency range below 200 Hz.

Aiming at the problem of low retrieval rate of basic contourlet transform texture image retrieval system, a texture image retrieval system based on Complex Contourlet Transform (CCT) was proposed. The contourlet transform used in the system was constructed by dual tree complex wavelet transform followed by critically subsampled directional filter banks, sub-bands energy and standard deviations were cascaded to form feature vectors, and the similarity metric was Canberra distance. The performance of retrieval system was compared including original contourlet transform, non-subsampled contourlet transform, semi-subsampled contourlet transform (CTS) and CCT under the same system structure. Experimental results show that the image retrieval system based on CCT has higher retrieval rate than those of CT, CTS and NSCT with almost same length of feature vectors, retrieval time and memory needed, and decomposition scale number can make significant effects on average retrieval rates.

A new structural frame of run-length associated indirectly with object was introduced to solve the problem of run-length labeling conflict more efficiently than current run-length connectivity analysis algorithms. An efficient object description algorithm based on such an improved run-length connectivity analysis was designed. The algorithm implemented run-length encoding through a successive scan, then proceeded to label run-length and merge the conflict objects by analyzing the connectivity with run-lengths in the above row. Meanwhile, the object description parameters were acquired such as area, perimeter, center of figure, Feret box. Both run-length region representation and object description parameters can be acquired synchronously just by such once scan. The algorithm has the advantage of low complexity and is easy to be implemented with high performance and deal with any complex figure.

Since the 4f optic information system composed of general equipments has not enough space invariant character, an improved method is presented to achieve inverse filtering denoising under the general condition. The step function response is obtained by taking advantage of the parallel processing of the 4f system and the linear theory is used to realize the inverse filtering denoising of the output image. Experiments show that the method can achieve the inverse filtering denoising, remarkably improve quality of the optic image and keep the detailed information under the current condition. Furthermore, the method doses not increase the complexity and time of the optic experiment. At the same time, it decreases the complexity and time of calculation of inversing filtering.

To avoid the disadvantage of the present algorithms of corner detection, which can only detect the locations of corners in images, but cannot pick up the angles of corners, an algorithm for picking up angles of corners in images is proposed based on Radon transform. The algorithm utilized SUSAN to detect corners, caught the neighborhood image of corners which were detected, and then judged whether the corner was measurable or not. If the corner was measurable, there was a grader mapping to each pixel. At last, Radon transform was used to extract edge lines of corners, and the geometry relation of coordinate was applied to pick up angles of corners. Radon transform could not detect original location of a line, which caused the defects of two values when one corner was picked up. To solve this problem, the same nucleus was used to angles picked-up through the idea of USAN for reference. Experiment results show the algorithm is effective.

To overcome the problem that the conventional algorithm based on Two-Dimensional Fisher Linear Discriminant (2DFLD) only took the row vectors of image matrix as sub-pattern, an ear recognition algorithm based on 2DFLD with taking the column vectors of image matrix as sub-pattern was proposed. Firstly, the ear feature subspace was extracted after processing training by using the column vectors of train image matrix as sub-pattern. Secondly, the test sample images were projected on small dimension subspace. Lastly, the nearest neighbor classifier to ear match based on Euclidean distance was used. The experimental results show that the recognition rate of column vectors reaches 98.333%, which is about 3.333% higher than that of row vectors. Compared with other methods such as PCA, 2DPCA and PCA+FLD based on the multi-element statistic analysis, the proposed method is the best one. It is an effective way of ear recognition.

In the classical CONDENSATION for object tracking, a prior transition probability, i.e., first or second order AR dynamic model is used to propagate the particles. However, it results in poor performance frequently. In order to propagate the particles efficiently, an improved CONDENSATION face tracking algorithm based on mean-shift drift is proposed. The approach uses the efficient mean shift tracker to attain coarse location of face target, then uses these results to determine the deterministic drift, finally propagates the particles with a small stochastic diffusion added. Because sampling via the proposed method can always make particles cluster around the true state region, the particles efficiency can be improved greatly. The experimental results of face tracking demonstrate that the performance of proposed algorithm is superior to the standard CONDENSATION.

To improve the performance of three-dimensional object recognition system and reduce computational complexity, a view-based method is proposed. First we extract color moments, texture features and affine invariant moments from the 2D view images of 3D objects. Color moments are robust and insensitive to the size and pose of objects. Texture features can distinguish objects which have similar shapes and different appearance. Affine invariant moments have the invariant properties under affine transformation. These features are combined into a feature vector of 23 elements, and then fed to Support Vector Machine (SVM) for training and recognition. We assessed our method based on two public 3D object dataset: COIL-100 and ALOI. 100% correct rate of recognition was obtained on both dataset when the number of presented training views for each object was 36 (10 degrees interval). When the number of training views was reduced, the correct rate of recognition was also satisfied and outperformed previous algorithms.