Abstract:The key technologies which based on the technology of image based distance measurement are researched for the avoidance of automobile collision. Firstly, according to the principle of relative motion and assume that the image distance measurement testbed is moved to collect images along the direction of the optical axis, and then the improved SIFT matching algorithm which is well used for image matching to obtain matching points. The world coordinates of the matching points are calculated to obtained distance value. Secondly, the validity of distance measurement principle is demonstrated by experiments. Finally, the image based distance measurement technology is applied into the automobile collision avoidance system. The error average of measurement system is 8.507 7 mm, the practicability and validity of the whole theme are proved by experiment results.

Abstract:In order to solve the effect of the image segmentation when the pedestrian image is collected by infrared camera, an image threshold segmentation algorithm based on adaptive particle swarm optimization of twodimensional OSTU is utilized. The gray scale of the current frame image and the neighborhood gray level of the current frame image pixel form a binary image. A 2D maximum betweencluster variance model is built up through calculating the average and variance between them, and combining with adaptive particle swarm optimization algorithm the best threshold image value is estimated. The algorithm can accurately estimate the threshold and reduce the calculation time. The simulation results demonstrate that the best image value is proper, the calculation time is shortened 50% when combine with adaptive particle swarm optimization algorithm. The proposed algorithm can get the optimal threshold quickly and accurately, and improve the segmentation effect of image preprocessing.

Abstract:In order to improve the instantaneity of the tracking method based on sparse and lowrank matrix decomposition, a lowrank object tracking algorithm based on large matrix and compressed feature is proposed in this paper. The matrix is decomposed sparsely and lowrankly by creating observation matrix using segmenting the large matrix into some parts. Then the error vector of each candidate object is obtained and an error matrix is built. The tracking result is gained by resolving the least 1norm of the error matrix. To adapt to the changes of target appearance, the dictionary is selectively updated based on the discrimination of vector similarity. When the tracking result is not trusted, it is updated by trajectory rectification. The instantaneity of the new algorithm is three times the old one via the comparison results on six typical sequences. The experiments demonstrate that the proposed algorithm can track the object accurately and robustly when there is part occlusion, illumination change and fast motion.

Abstract:To provide a better image deblurring and remove successfully the speckle noise, a novel PDEbased image denoising approach is proposed in this paper. It is based on a nonlinear fourthorder diffusion model. The nonlinear PDE scheme is described first. Then, a mathematical treatment is provided for this differential model, its wellposedness being investigated. It is proved that the model is wellposed in some certain conditions and admits a weak solution. The weak solution of the obtained PDE is approximated by developing an explicit finitedifference based numerical discretization scheme. The experimental results show that the new model proposed in this paper can achieve good results in image denoising and preserving edges and other details. Compared with the classical model, the peak signal to noise ratio is greatly improved and the denoising performance is more better.

Abstract:A method for insulator location and selfexplode detection is designed. The recognition and location algorithm firstly uses the Otsu method to segment the insulator, the moment invariants of insulators are extracted. Finally, Adaboost classifier is used to locate the insulator. According to the shape features of selfexplode insulator, a detection method is designed by calculating Euclidean distance of adjacent insulators. The selfexplode detection method works well in dealing with multiple explosion point. The method accurate rate reached 87%. Through the experiment, the accuracy of this method is better, and it is more suitable for practical application.

Abstract:Aiming at overcoming the problems that the traditional modal testing methods have limited information and the transducer additional mass effects modal parameters, and existing noncontact modal testing method requires complex image processing, a vibration modal measurement method for thinwalled parts using optical flow point matching and tracking is proposed. It avoids complicated image processing such as feature segmentation and extraction and so on in each frame. Firstly, the model of vibration detection based on the inplane vision is developed. The intrinsic and extrinsic parameters of industrial camera are calibrated through the camera calibration board, and the errors are analyzed. The principle and the method of monocular vision vibration testing based on optical flow method are investigated. The vibrating sequence images of structure which has pasted feature points are captured with a monocular industrial camera. The pyramid LucasKanada algorithm is used for optical flow point matching and tracking, and to get the vibration information of each subpixel feature point. And then the modal parameters are obtained by modal parameter identification. Based on the proposed method, a thinwalled vibration modal testing system is built and the vibration modal testing experiments with the thinwalled beam is investigated, the testing results are compared and analyzed between that of a shaker testing and FEM simulation. The results show that the natural frequency errors are less than 5% and the corresponding vibration modes are the same, which verifies the correctness of the proposed method, so the proposed method provides a new way for the vibration modal testing of thinwalled parts.

Abstract:Map merging is a crucial technology for cooperative mapping of multirobot system. A multirobot grid map merging method based on SURF (speeded up robust features) algorithm is studied. In the approach, the robot motion coordinate system is transformed into a rigid body. The mathematical model of the grid map merging problem is establishmented as minimization problem of image registration. Firstly, an improved SURF algorithm is used to extract the features from the grid maps. Secondly, the RANSAC (random sampling consensus) algorithm is employed to eliminate the mismatch and get the initial merging parameters, and the merging parameter is used as the initial value of the ICP (iterative nearest point) algorithm to solve the objective function. Finally, the open datasets and the Turletbot2 mobile robots are adopted for experiments. The experimental results verify that the proposed method can achieve good robustness and high accuracy in the grid map merging by multirobots, the fast speed of map merging indicates that it is suitable for large scale environment.

Abstract:In order to improve the watermarking robustness and solve the contradiction between the transparency and robustness of the watermarking, a digital image watermarking algorithm in wavelet domain based on singular value decomposition (SVD) and human visual system (HVS) is proposed. Firstly, the carrier image is carried on twolevel discrete wavelet transform (DWT), and its second level low frequency subband is divided into blocks. Then, each subblock is carried on SVD, the modified cosine similarity measurement method is used to calculate the best watermarking embedding position according to HVS sensitivity to the average brightness. Finally, the encrypted watermarking is embedded into the selected singular values by the quantizing way, and the embedding strength is adaptively adjusted according to singular value of the carrier image subblock. The experimental results show that the transparency of the watermarking is very good, and it has high robustness against noise, filtering, cropping, JPEG compression and so on.

Abstract:In order to solve the problem such as need the strict optical calibration in current optical image encryption algorithm, the optical image encryption algorithm based on circular harmonic component expansion and Gyrator transform domain phase retrieval was proposed in this paper. Firstly, Gyrator transform spectrum was formed by introducing Gyrator transform to deal with the plain. Then the spectrum of Gyrator transform is divided into zero order harmonic components and nonzero order harmonic components based on offaxis circular harmonic component expansion mechanism. The cipher was obtained by introducing the spherical phase factor to modulate the nonzero order harmonic component. Finally, the complex distribution was obtained by using the retrieval algorithm of Gyrator transform to take the zero order and nonzero order amplitude as the constraints condition of input plain and output cipher to finish the image encryption. The experimental results show that this algorithm has higher security and anti filtering robustness with stronger antinoise attack and antishear attack ability. This proposed algorithm can better protect the image in the network security transmission with good practicality.

Abstract:Aiming at the problem of steady state voltage stability caused by distributed generation access to distribution network, the voltage stability of active distribution network including distribution network is mainly investigated in the paper. To reflect system voltage stability of each node, a voltage stability index method is firstly proposed according to the results of power flow calculation for the whole distribution network. Considering distributed generation as a negative load direct injection, different models of distributed generation in power flow calculation are then established. Finally, taking the typical wind power of distributed generation for example, active distribution network of IEEE33 node is simulated, and different fan efforts, fan access and active output fluctuation influence on the voltage stability of active distribution network system are analyzed in detail. The simulation results show that the distributed network with distributed generation can effectively improve the voltage stability.

Abstract:Rolling bearing is significant research content for rolling machine condition monitoring and fault diagnosis. In order to diagnose the rolling bearing fault position and degree more effectively, a rolling bearing fault diagnosis method based on Hilbert marginal spectrum and support vector data description (SVDD) optimized by improved particle swarm optimization (IPSO) is proposed. In this method, the rolling bearing vibration signal is decomposed into a set of intrinsic mode functions (IMFs), then marginal spectrum and autoregressive (AR) model parameters are established and system feature vector is constructed of AR parameters and feature power function, which is obtained from marginal spectrum. In order to solve the problem of deciding SVDD’s significant parameters by traditional gridsearching or experience, a method using dynamic factor based particle swarm algorithm is used to find the optimized SVDD’s significant parameters penalty constant C and kernel function width σ, and the optimized model is put into use of intelligent rolling bearing fault diagnosis. The experiment results of manual and real data sets show that different kinds of rolling bearing fault conditions can be recognized effectively by the proposed method with higher efficiency and precision than traditional gridsearching method.

Abstract:There is a large number of line traces on the surface of the broken end which left in the cable cutting case crime scene of highspeed railway, they often present nonlinear morphological feature and has strong randomness. In order to make rapid trace analysis and infer the tools, a fast algorithm based on wavelet domain feature aiming at the nonlinear line traces is presented. The proposed algorithm applies wavelet decomposition to the 1D signal which picked up by single point laser displacement sensor to partially reduce noises. After that, the dynamic time warping is employed to realize the trace feature similarity matching. Finally, the linear regression machine learning algorithm based on gradient descent method is used to realize the constant iteration. The experiment results of cut traces sample data comparison demonstrate the accuracy and reliability of the proposed algorithm.

Abstract:Multitarget tracking is very important in intelligent video surveillance system. Occlusions among many moving objects and objects with similar appearance that could result in incontinuous trajectory are challenging problems in this field. Based on the fact that the relative positions remain stable between two close pairs, a novel multitarget tracking algorithm based on conditional random field is presented in this paper. Unlike previous approaches which focus only on appearance and motion models for all targets, we consider discriminative features for distinguishing difficult pairs of targets with sets of labels. Multilevel is adopted by this means, in which the set of tracklets produced by previous level is used as an input. And tracking problem is transformed into an energy minimization problem including a set of unary function for a continuous trajectory and a set of pairwise function for pairs of tracklets. This new method is more powerful in deal with objects with similar appearance and badly occlusions compared to stateofart methods. Qualitative and quantitative experimental results show that this new method has a better performance.

Abstract:In order to solve the target recognition and classification problem of battlefield acoustic target detection system, an energy feature analysis (EFA) method based on cutoff frequency EEMD (CFEEMD) is proposed. Selecting the minimum effective frequency of the signal itself as screening termination condition of EEMD, the EEMD method is improved to achieve rapid decomposition of acoustic target and get accurate IMF components. The total energy vector of the target signal is obtained by calculating the energy of each IMF, and then the energy distribution of each IMF component of the typical target acoustic signal is analyzed. The energy difference between the high and low frequency of the target acoustic signal is defined, which is used as feature parameter to identify and classify the battlefield acoustic target. Through the semiphysical simulation experiment the feasibility and the practicality of the EFAbased target recognition method with improved EEMD is verified, which is suitable for identification and classification of battlefield acoustic target.

Abstract:Aiming at the problem that the low SNR speech endpoint detection accuracy is seriously affected by the background, a speech endpoint detection method based on projection classification is proposed in this paper. Firstly, the phonetic characteristics of low SNR environment is calculated using long speech signal rate measure characteristics. The method makes full use of different speech signal and a voice signal to enhance the degree of differentiation of low SNR condition. Secondly, by using Fisher criterion, the classification identification of the voice and the background noise is carried out to ensure that the projection parameters have the smallest similar characteristics and the largest different characteristics. The experimental results show that the proposed method has high detection accuracy, the correct detection rate is more than 86.7% even in the SNR=-10 dB white noise interference condition.

Abstract:To solve the problem of premature convergence in traditional particle swarm optimization (PSO) and differential evolution (DE), a hybrid algorithm based on particle swarm optimization with stochastic differential mutation is proposed in this paper. Combining with the characteristics between PSO and DE, the new algorithm firstly generates a candidate individual using differential mutation, and then put the individual into velocity update formula leading flight direction of particle, which can expand the search space and enhance the global explorative ability of algorithm. Meanwhile, a stochastic differential mutation method is presented to disturb the current optimal particle in order to avoid the best particle being trapped into local optima, since which may cause search stagnation. The new algorithm compared with three related algorithms on 8 benchmark functions including unimodal and multimodal test functions. The experimental results show that the new hybrid algorithm outperforms other comparative algorithms and greatly improves performance of algorithm.

Abstract:The RFID threedimensional localization algorithm is the main technology of indoor localization. Aiming at the problems of low location accuracy and poor adaptability in the traditional LANDMARC localization algorithm, a RFID 3DLANDMARC localization algorithm based on the cultural double quantum particle swarm optimization is proposed. Firstly, the advantages of the BP neural network in data fitting is used to preprocess the acquired signal and the wireless signal transmission loss model is studied to improve localization accuracy of LANDMARC algorithm. With the purpose of solving the adaptive problem existed in LANDMARC localization algorithm, the cultural double quantum particle swarm optimization (CDQPSO) algorithm is introduced, which has the technology advantages in global search and optimization to solve the localization model. The experimental results show that the proposed algorithm improves the localization accuracy and adaptability significantly, compared with the basic LANDMARC algorithm and particle swarm optimization LANDMARC algorithm, with the localization error of 75% of tested label is less than 0.56 m, and it can overcome the shortcoming of slow convergence existed in particle swarm optimization.

Abstract:Unmanned aerial vehicles (UAVs) equipped with HD camera can obtain a large number of detailed inspection images of the insulators which are indispensable components in the transmission lines. A quick and accurate detection of the insulator can provide a reliable basis for distance measurement and obstacle avoidance for UAV when flying close to towers for details. Simultaneously, as a faultprone component, the insulators seriously threat the network security, thus computer technology is required for fault diagnosis. The detection of the insulator image with the complex aerial background is implemented by constructing a convolutional neural network (CNN), which has the classic architecture of five modules of convolution and pooling, two modules of fully connected layers. In this paper, a recognition algorithm for explosion fault based on saliency detection is proposed, which uses the trained network model to extract the features. Then we put the saliency maps into a SOM network and build the mathematical module via super pixel segmentation, contour detection, and other image processing methods. The test shows that the algorithm can reduce the errors caused by manual analysis. The test also demonstrates that the detection rate of the insulator can reach 90% and the recognition rate of explosion fault can reach 85% with complex background. They effectively improve the efficiency of inspection and make the inspection more intelligent.

Abstract:The control of vibration amplitude of flow tube is very essential for Coriolis mass flowmeter (CMF) in the application of measurement. When the singlephase flow and gasliquid twophase flow switch to each other frequently, the current control method would lead to a large attenuation and overshoot of vibration amplitude. Therefore, the experimental scheme is designed and conducted. The reason of the problem is analyzed, and then the variable drive cycle method and the variable integrallimitingvalue method are proposed. Two methods are realized in real time on the developed digital mass flow transmitter and verified experimentally. The experimental results show that the variable drive cycle method reduces the attenuation degree of the vibration amplitude effectively when the singlephase flow switch to the gasliquid twophase flow, and the variable integrallimitingvalue method solves the significant overshoot of the vibration amplitude well when the gasliquid twophase flow switch to the singlephase flow.

Abstract:Aiming at the shortcoming of low accuracy and poor realtime performance of voltage sag detection method, a new method based on improved dq transform and sequential morphological filter is proposed. Basic principle of improved dq transform for voltage sag detection algorithm is researched, the concept of order morphology is introduced. And thus, the morphological filter is designed, and the method of selecting filter parameter is expounded. By setting different signal noise ratio and leading angle, the simulation results show that the new method can detect the sag under ideal condition and nonideal condition. It is proved that the new method can detect the eigenvalues and the stopping time of the sag more efficiently and accurately.

Abstract:In application of active power filters (APF), DC side voltage fluctuation is very common and the secondary pulsation is particularly prominent in this kind of fluctuation. In the study of threephase threewire APF with asymmetric load, it is found that voltage loop is affected by secondary pulsation in some degree and it will generate unnecessary additional negative sequence reference current and reference current amplification effect. As a result, the fundamental negative sequence current in compensation current produced by APF is greater than fundamental negative sequence current produced by asymmetrical load, which causes the asymmetry of the network side current. After secondary pulsant suppressed by notch filter, the reference current amplification effect can be reduced effectively and the network side current will be symmetrical. The simulation and the experiments have proved the correctness of views proposed by this paper.

Abstract:Virtual flux has the advantages of simple calculation and easy to digital operation, which has attracted much attention in the application of alternating measurement. But due to the defect of the calculation, the directcurrent bias and the integrator saturation phenomenon caused by the integration of the grid voltage are seriously affecting the space vector orientation. In the process of the voltage measurement, the control method is added to the virtual flux technology. A virtual flux observer is designed, which is used to replace the traditional PhraseLock technology. The directcurrent bias and the initial integration error are effectively eliminated. Besides, the virtual flux optimization algorithm is derived, and the extended Kalman filtering in imitation system is used to imitate the sampling voltage. The method can effectively reduce the harmonic distortion and the time of single cycle control. The simulation of the observer and the algorithmic demonstration show that the observer optimized by algorithmic can eliminate the phenomenon of integral saturation, and the influence of directcurrent bias on the precision orientation, and improve the dynamic performance and control precision of the system.