In order to get the horizontal angle of the head of full face tunnel boring machine, a fiber optic gyroscope strap-down north-seeker is proposed, effectively avoiding the problems of bad visibility and smoke interference when using laser guiding system. The horizontal component of the earth's angular velocity is measured by fiber optic gyroscope, and elevation angle and roll angle of the tunnel boring machine is measured by inclinometer. Tilt error of gyroscope sensitive axis is compensated by coordinate rotation transformation. Moreover, latitude is eliminated by structural formulas using the components of the earth's angular velocity in the carrier coordinate. The horizontal angle of the tunnel boring machine can be calculated by four-position algorithm. The strap-down north-seeker is installed in a full face tunnel boring machine and can achieve 0.07° seeking north precision in the experiment of working condition, which satisfies the requirement of tunnel construction.

Icing identification technology of optical fiber composite overhead ground wire (OPGW) is studied based on fiber Brillouin scattering. Thermodynamic theoretical derivation of identification technology is given, and then temperature distribution of conductor is simulated with the finite element method by Comsol Multiphysics. And the results show that icing area is dull to the change of ambient temperature. Hence ice-coating on transmission lines can be identified through temperature variation in a period of time. Based on the results, OPGW icing identification system based on fiber Brillouin scattering is designed using Brillouin Optical Time Domain Reflectometer, which solves problems of insufficient power and electromagnetic interference in traditional identification systems. This system is successfully applied to the Zhaotong power system with detection range 75 km and temperature error ±2 ℃.

To effectively solve the problem of fast and accurate detection of moving targets in complex surveillance scene, target detection method based on multi-frame background subtraction and Cauchy model is proposed. Firstly, Surendra background model is improved to get clean background image. Then, system judges the current pixel on the absolute differential image belonging to the target areas or background areas by the absolute difference between the current background frame and the real-time video frame. Finally, through the Cauchy distribution model of the pixel, the aim of the moving target detection is realized in complex surveillance scene. The experiment on the vehicle, pedestrian and other object shows that the method can not only suppress the noise and interference of false target, but also can segment foreground target rapidly and accurately.

In order to solve tracking problem of maneuvering target in nonlinear background, an Augmented Interacting Multiple Model Cubature Kalman Filter (AIMMCKF) algorithm is put forward. To obtain the fixed-lag smoothing state estimation, IMMCKF approach is applied to a nonlinear state-augmented system in the proposed algorithm. At the same time, to tackle different models problem within being represented in different state spaces, corresponding augmented conversion operation can be used. The simulation results show that, the proposed algorithm achieves higher precision and stronger adaptability for maneuvering target tracking in comparison with traditional nonlinear tracking algorithms.

In order to realize the high precision moving object detection in snowy weather condition, this paper improves the GMM algorithm. Firstly, the multi-resolution, high and low threshold concepts are used to optimize the detection results, which can overcome the influence of the dynamic background noise. Then, the color model is used to suppress the weak shadows and illumination changes by moving objects. Finally, the hole is filled in the rectangle with the minimum constraint of each object, and the motion mask is filled due to the excessive suppression of the shadow and the loss of the surface covered by snow. Experimental results show that the improved algorithm is better than the GMM algorithm for all of the seven indicators. Compared with the current outstanding algorithm FTSG, there are four of the seven transcend, the two close.

Star sensor is a kind of space attitude measurement device whose accuracy reaches arc-second degree, and its imaging module consists of optical lens and image sensor which mainly refers to CMOS APS. The assembly of APS of star sensor is focused, and the influence of spatial position and attitude parameters of APS on the calculation of observation vector is derived theoretically. Thus, a method to control and calibrate these parameters during the process of assembly is proposed, and a corresponding physical system is developed. Test data shows that using this system, the calibration repeatability errors of two components of tilt are ±1.05″ and ±1.09″, respectively, the calibration repeatability error of roll is ±9.4″, and the calibration repeatability errors of two components of eccentricity are ±0.53 μm and ±0.55 μm, respectively.

According to the imaging process of star over saturation phenomenon, the characteristics of Point Spread Function (PSF) of star image points are analyzed, and the energy distribution model of over saturation star image is established, and a PSF correlation algorithm is proposed. The algorithm first identifies the PSF sequence with the best star regional similarity, positioning by finding the maximum value of the sequence. Simulation experiments show that the positioning error of the PSF correlation method is as low as 0.05 pixels under the condition that the star image points are noisy. Under the asymmetric PSF condition, the positioning error is still less than 0.1 pixels, which is higher than the traditional centroid method, surface fitting method and other methods.

According to the characteristics of type-B ultrasonic image with low contrast and SPECT image with low spatial resolution, an image fusion algorithm based on Nonsubsampled Shearlet Transform (NSST) and improved Pulse Coupled Neural Network (PCNN) is proposed. The NSST is used to decompose two registered source images, and low frequency sub-band coefficients and high frequency sub-band coefficients with different scales and directions are obtained. Low frequency coefficients are fused by the maximum of the regional energy. High frequency coefficients are fused by improved PCNN algorithm. The Sum Modified Laplacian is used for the input of PCNN, and the Energy of Gradient is used for the link intensity of PCNN, thus the high-frequency coefficients are selected by the sum of ignition output amplitude maximum. Finally, the fused image is reconstructed by inverse NSST. Experimental results demonstrate that the proposed algorithm achieves good results in the subjective perspective and objective criteria.

The cable harness is widely used in electronic equipment and wiring component. Its quality can be measured by analyzing the length of wire inside diameter and outside diameter. The level set method is applied to segment the microscopic image of cable harness. To deal with adjoining object, we reset the RGB channels of the image and convert them to Lab space, where the level set method is applied to segment the inner counter. Lab space is selected for inner contour segmentation as it can enhance the difference between blue and yellow color information, which is an essential feature for this kind of microimage. Finally, the binary image is processed by mathematical morphology technique before obtaining the final internal contour. The experimental results show that the proposed method can obtain contours efficiently and make it feasible to the data extraction and quality control.

To avoid the adverse effects of edge burrs and the difference of contrast ratio, a corner detection algorithm via multi-scale differential mode similarity is proposed. Based on the edge contour extraction, it extracts the local differential mode at each edge pixel by using the Anisotropic Gaussian Directional Derivative (ANDD) filters. Considering the fact that the differential mode at single pixel suffers from edge burrs, thus the similarity between the differential modes at a pixel and its support region is constructed as corner measure to enhance the stability of the corner measure. Finally, the measures at three scales are fused into a new corner measure to enhance the noise robustness and to improve the localization accuracy of corner measure. The experimental results show that the proposed algorithm behaves better performance in detection accuracy ratio, false detection ratio, localization and repeatability compared with three state-of-art algorithms.

For unstructured road image, a road detection and intelligent warning algorithm is proposed based on vanishing point information. Segmentation algorithm firstly employed the straight line information, vertical envelope, grouping strategy and path clustering algorithm to estimate the vanishing point. Then soft voting method is proposed to extract road region based on the boundaries and vanishing point information features. Finally, according to the features of unstructured road region, the navigation line is extracted, and yaw distance and yaw time warning decisions are depended on the vehicle yaw Angle. Thus, the intelligent lane departure warning algorithm can be achieved. Experimental results show that the proposed algorithm can quickly and efficiently extract the road region of unstructured road. At the same time, the algorithm can be effective for lane departure warning according to early warning parameters.

In order to comprehensive utilize infrared and visible imagery weather information, a kind of infrared and visible light satellite cloud image fusion method is put forward based on Nonsubsampled Shearlet Transform (NSST) and adaptive Pulse Coupled Neural Network (PCNN). Firstly, the infrared and visible satellite imagery were decomposed at multi-scale and multi-direction by NSST, then for the low frequency subband coefficients, an self-adaptive fusion rule algorithm based on local area energy and local area variance was presented. The high frequency subband coefficients are fused by an improved adaptive PCNN, The connection strength of pulse coupled neural network is determined by a S type fuzzy membership function according to the different importance of the regional features of high frequency coefficients. Finally, the fusion of low frequency and high frequency were reconstructed by NSST inverse transform. Experimental results show that the proposed method of image fusion is better than the typical fusion method of comparison in this paper both from subjective visual effect and objective evaluation index, and fusion cloud image can provides more rich meteorological data of weather information for the subsequent weather analysis and processing.

For the video stream copyright issues of high efficiency video coding (HEVC), a new zero-watermarking algorithm with robustness to re-quantization transcoding is proposed. Firstly, from statistics analysis about re-quantization transcoding, it is found that Coding Unit (CU) depths have strong stability and only a fraction of the depths would shift and almost shift to adjacent depths. Then, in order to increase the robustness of the depth characteristic, the CU depths are divided into two groups and mapped into two values ‘0’ and ‘1’. Finally, ‘xor’ operation is performed between the binary information encrypted and the copyright information scrambled by using the chaotic algorithm, the outcome with the timestamp acts as the ultimate registered zero-watermarking. Experimental results show that the proposed algorithm has strong robustness to re-quantization transcoding and other common signal attacks.

In video motion detection, moving shadow is often mistakenly detected as moving objects. In order to solve this problem, a moving shadow detection method is presented, which uses Haar Local Binary Pattern (HLBP) to detect shadow. Firstly, HLBP feature of detected moving regions and corresponding regions in the background is extracted. Secondly, it uses Manhattan Distance to measure the difference of HLBP feature between the two regions to get a map about texture difference. Finally, Ostu algorithm is employed to segment texture difference map to get a mask of moving objects, and then moving objects are extracted. The experimental results show that the proposed method can effectively detect moving shadow in some outdoor and indoor scenes in real time.

A novel method is proposed to study optical imaging system using the information theory. The light intensity function is decomposed into linearity combination of a series of harmonic waves without negative value by improved Fourier transform. The normalized harmonic wave coefficient set of object is regarded as information source and that of image is regarded as information home. The optical information channel matrix is obtained by eigenvalue of imaging integral equations. Moreover, average mutual information and information channel capacity are calculated that can evaluate the information transmission ability of the optical system.