To address the challenge of dynamic changes in object relationships over time in videos， a video dynamic scene graph generation model based on multi-scale spatial-temporal Transformer was proposed. The multi-scale modeling idea was introduced into the classic Transformer architecture to precisely model dynamic fine-grained semantics in videos. First， in the spatial dimension， the attention was given to both the global spatial correlations of objects， similar to traditional models， and the local spatial correlations among objects’ relative positions， which facilitated a better understanding of interactive dynamics between people and objects， leading to more accurate semantic analysis results. Then， in the temporal dimension， not only the traditional short-term temporal correlations of objects in videos were modeled， but also the long-term temporal correlations of the same object pairs throughout the entire videos were emphasized. Comprehensive modeling of long-term relationships between objects assisted in generating more accurate and coherent scene graphs， mitigating issues arising from occlusions， overlaps， etc. during scene graph generation. Finally， through the collaborative efforts of the spatial encoder and temporal encoder， dynamic fine-grained semantics in videos were captured more accurately by the model， avoiding limitations inherent in traditional single-scale approaches. The experimental results show that， compared to the baseline model STTran， the proposed model achieves an increase of 5.0 percentage points， 2.8 percentage points， and 2.9 percentage points in terms of Recall@10 for the tasks of predicate classification， scene graph classification， and scene graph detection， respectively， on the Action Genome benchmark dataset. This demonstrates that the multi-scale modeling concept can enhance precision and effectively boost performance in dynamic video scene graph generation tasks.

The traditional Convolutional Neural Network (CNN) can only extract local features for human behaviors and actions, which leads to low recognition accuracy for similar behaviors. To resolve this problem, a two-stream Non-Local Residual Network (NL-ResNet) based behavior recognition method was proposed. First, the RGB (Red-Green-Blue) frame and the dense optical flow graph of the video were extracted, which were used as the inputs of spatial and temporal flow networks, respectively, and a pre-processing method combining corner cropping and multiple scales was used to perform data enhancement. Second, the residual blocks of the residual network were used to extract local appearance features and motion features of the video respectively, then the global information of the video was extracted by the non-local CNN module connected after the residual block, so as to achieve the crossover extraction of local and global features of the network. Finally, the two branch networks were classified more accurately by A-softmax loss function, and the recognition results after weighted fusion were output. The method makes full use of global and local features to improve the representation capability of the model. On UCF101 dataset, NL-ResNet achieves a recognition accuracy of 93.5%, which is 5.5 percentage points higher compared to the original two-stream network. Experimental results show that the proposed model can better extract behavior features, and effectively improve the behavior recognition accuracy.

In most speech separation and speech enhancement algorithms based on deep learning, the spectrum feature after Fourier transform is used as the input feature of the neural network, without considering the phase information in the speech signal. However, some previous studies show that phase information is essential to improve speech quality, especially at low Signal-to-Noise Ratio (SNR). To solve this problem, a speech separation algorithm based on Convolutional Encoder Decoder network and Gated Recurrent Unit (CED-GRU) network was proposed. Firstly, based on the characteristic that the original waveform contains both amplitude information and phase information, the original waveform of the mixed speech signal was used as the input feature. Secondly, the timing problem in speech signal was able to be effectively solved by combining the Convolutional Encoder Decoder (CED) network and the Gated Recurrent Unit (GRU) network. Compared with Permutation Invariant Training (PIT) algorithm, DC (Deep Clustering) algorithm, Deep Attractor Network (DAN) algorithm, the improved algorithm has the Perceptual Evaluation of Speech Quality (PESQ) and Short-Time Objective Intelligibility (STOI) of men and men, men and women, women and women increased by 1.16 and 0.29, 1.37 and 0.27, 1.08 and 0.3; 0.87 and 0.21, 1.11 and 0.22, 0.81 and 0.24; 0.64 and 0.24, 1.01 and 0.34, 0.73 and 0.29 percentage points. The experimental results show that the speech separation system based on CED-GRU has great value in practical application.

In order to solve the index calculation problems of high complexity, strong completeness and low efficiency in the filed of financial foreign exchange market monitoring, a novel distributed computing framework for foreign exchange market monitoring based on Spark big data structure was proposed. Firstly, the business characteristics and existing technology framework for foreign exchange market monitoring were analyzed and summarized. Secondly, the foreign exchange business features of single-market multi-indicator and multi-market multi-indicator were considered. Finally, based on Spark's Directed Acyclic Graph (DAG) job scheduling mechanism and resource scheduling pool isolation mechanism of YARN (Yet Another Recourse Negotiator), the Market-level DAG (M-DAG) model and the market-level resource allocation strategy named M-YARN (Market-level YARN) model were proposed, respectively. The experimental results show that, the performance of the proposed distributed computing framework for foreign exchange market monitoring improves the performance by more than 80% compared to the traditional technology framework, and can effectively guarantee the completeness, accuracy and timeliness of foreign exchange market monitoring indicator calculation under the background of big data.

In order to solve the problem that the traditional cascaded Convolutional Neural Network (CNN) has low accuracy of feature point localization in left ventricular ultrasound image, an improved cascaded CNN with region extracted by Faster Region-based CNN (Faster-RCNN) model was proposed to locate the left ventricular endocardial and epicardial feature points in ultrasound images. Firstly, the traditional cascaded CNN was improved by a structure of two-stage cascaded. In the first stage, an improved convolutional network was used to roughly locate the endocardial and epicardial joint feature points. In the second stage, four improved convolutional networks were used to fine-tune the endocardial feature points and the epicardial feature points separately. After that, the positions of joint contour feature points were output. Secondly, the improved cascaded CNN was merged with target region extraction, which means that the target region containing the left ventricle was extracted by the Faster-RCNN model and then was sent into the improved cascaded CNN. Finally, the left ventricular contour feature points were located from coarse to fine. Experimental results show that compared with the traditional cascaded CNN, the proposed method is much more accurate in left ventricle feature point localization, and its prediction points are closer to the actual values. Under the root mean square error evaluation standard, the accuracy of feature point localization is improved by 32.6 percentage points.

A time series trend prediction algorithm at multiple time scales based on novel feature model was proposed to solve the trend prediction problem of stock and fund time series data. Firstly, a feature tree with multiple time scales of features was extracted from original time series, which described time series with the characteristics of the series in each level and relationship between levels. Then, the hidden states in feature sequences were extracted by clustering. Finally, a Multiple Time Scaled Trend Prediction Algorithm (MTSTPA) was designed by using Hidden Markov Model (HMM) to simultaneously predict the trend and length of the trends at different scales. In the experiments on real stock datasets, the prediction accuracy at every scale are more than 60%. Compared with the algorithm without using feature tree, the model using the feature tree is more efficient, and the accuracy is up to 10 percentage points higher at a certain scale. At the same time, compared with the classical Auto-Regressive Moving Average (ARMA) model and pattern-based Hidden Markov Model (PHMM), MTSTPA performs better, verifying the validity of MTSTPA.

To improve the convergence speed of distributed online optimization algorithms, a fast first-order Online Distributed Dual Average optimization (FODD) algorithm was proposed by sequentially adding edges to the underlying network topology. Firstly, aiming at solving the problem of the online distributed optimization to make the selected edge and network model mix quickly by using the method of edge addition, a mathematical model was established and solved by FODD. Secondly, the relationship between network topology designed and the convergence rate of the online distributed dual average algorithm was revealed, which clearly showed that, by improving the algebraic connectivity of the underlying topology network, the Regret bound could also be greatly improved. The Online Distributed Dual Average (ODDA) algorithm was extended from static networks to time-varying networks. Meanwhile, the proposed FODD algorithm was proved to be convergent and the convergence rate was specified. Finally, the results of numerical simulations show that, compared with existing algorithms such as ODDA, the proposed FODD algorithm has better convergence performance.

Aiming at the imbalance of training data set, an integrated support vector machine classification algorithm was proposed by combining sampling method with ensemble method. Firstly, unsupervised clustering was performed on an unbalanced training set, then the underlying local attention support vector machine was used to partition the data set so as to precisely control the local features of data sets. Finally, top support vector machine was used to predicte classification. The evaluation results on UCI dataset show that compared with the popular algorithms such as sampling based Kernelized Synthetic Minority Over-sampling TEchnique (K-SMOTE), integration based Gradient Tree Boosting (GTB) and cost sensitive ensemble algorithm (AdaCost), the proposed support vector machine algorithm can significantly improve the classification effect and solve the problem of unbalanced data set to a certain extent.

To solve the problems of local or global sparse representation based image saliency detection methods, such as incomplete object extracted, unsmooth boundary and residual noise, an image saliency detection algorithm based on adaptive fusion of local sparse representation and global sparse representation was proposed. Firstly, the original image was divided into a set of image blocks, and these blocks were used to substitute the image pixels, which may decrease the computational complexity. Secondly, the blocked image was represented via local sparse representation. Specifically, for each image block, an overcomplete dictionary was generated by using its surrounding image blocks, and based on such dictionary the image block was sparsely reconstructed. As a result, an initial local saliency map which may effectively extract the edges of the salient objects could be gotten. Thirdly, the blocked image was represented by global sparse representation. The procedures were similar to the above steps. The difference was that, for each image block, the overcomplete dictionary was constructed by using the image blocks from the four margins of the input image. According to this, an initial global saliency map which could effectively detect the inner areas of the salient objects was obtained. Finally, the initial local and global saliency maps were adaptively fused together to compute the final saliency map. Experimental results demonstrate that compared with several classical saliency detection methods, the proposed algorithm significantly improves the precision, recall and F-measure.

In order to analyze the effect of the correlation between the last-mile delivery and time slot existing in the customer choice procedure of urban distribution service on the operational decisions, such as reception box locating, time slot allocating and vehicle routing, a nested Logit model was used to quantify the customer's choice of delivery service options, and a two-tier nested Logit selection model for urban delivery was proposed. Then a multi-objective optimization model integrated with reception box locating, time slot allocating and vehicle routing was constructed in the purpose of maximizing the delivery amount and minimizing the delivery cost. At last, a Multi-Objective Particle Swarm Optimization (MOPSO) algorithm was constructed to solve this model based on non-dominance sorting, adaptive grid and crowding distance sorting. The analysis shows that, as the attended-home-delivery independence parameter is gradually increased, the substitution of customer demand in different time slots is smaller, the optimal solutions tend to improve the delivery punctuality with increasing the delivery amount, whether it is to minimize the cost or to maximize the number of delivery; on the contrary, with the increase of the reception-box independence parameter, the optimal solutions will decrease the delivery punctuality with reducing the delivery amount.

In order to explore the possibility of using two microphones for separation and locating of multiple sound sources in a two-dimensional plane, an indoor voice separation and sound source localization system based on dual-microphone was proposed. According to the signal collected by microphones, a dual-microphone time delay-attenuation model was established. Then, Degenerte Unmixing Estimation Technique (DUET) algorithm was used to estimate the delay-attenuation parameters of model, and the parameter histogram was drawn. In the speech separation stage, Binary Time-Frequency Masking (BTFM) was established. According to the parameter histogram, binary masking method was combined to separate the mixed speech. In the sound source localization stage, the mathematical equations for determining the location of sound source were obtained by deducing the relationship between the model attenuation parameters and the signal energy ratio. Roomsimove toolbox was used to simulate the indoor acoustic environment. Through Matlab simulation and geometric coordinate calculation, the locating in the two-dimensional plane was completed while separating multiple targets of sound source. The experimental results show that, the locating errors of the proposed system for multiple signals of sound source are less than 2%. Therefore, it contributes to the research and development of small system.

Concerning the problem that most of the existing social network growing models have negative degree correlation, considering the characteristics of positive degree correlations and high clustering coefficients, a new social network growing model was proposed based on Holme and Kim (HK) model. Firstly, the topological structure of a real-world social network was analyzed to obtain some important topological parameters of real social networks. Secondly, the HK model was improved by introducing triad formation mechanism, namely HK extended model with Turnable Degree Correlation and Clustering coefficient (HK-TDC&C), by which both clustering coefficients and degree correlations in the network could be adjusted. The model could be used to construct social networks with various topological properties. Finally, using mean field theory, the degree distribution of the model was analyzed, and Matlab was used for numerical simulation to calculate other topological parameters of the network. The results show that, by turning preferred attachment parameters and connection probabilities, the social network constructed by HK-TDC&C model can satisfy the basic characteristics of social networks, including scale-free characteristics, small world characteristics, high clustering coefficient characteristics and degree positive correlation properties, and its topology is closer to the real social network.

A large amount of alarms considerably complicate the root-cause analysis in telecom networks, thus a new alarm filtering algorithm was proposed to minimize the interference on the analysis. Firstly, a quantitative analysis for the alarm data, e.g., the quantity distribution and the average duration, was conducted, and the concepts of alarm impact and high-frequency transient alarm were defined. Subsequently, the importance of each alarm instance was evaluated from four perspectives:the amount of the alarms, the average duration of the alarms, the alarm impact, and the average duration of the alarm instance. Accordingly, an alarm filtering algorithm with O ( n) computation complexity in principle was proposed, where n is the number of alarms under analysis. Single-factor experimental analysis show that the compression ratio of the alarm data has a positive correlation with the alarm amount of a specific alarm element, the average duration of the alarms, the alarm impact, and the duration of the alarm instance; further, the accuracy of the proposed algorithm is improved by 18 percentage points at most compared with Flexible Transient Flapping Determination (FTD) algorithm. The proposed algorithm can be used both for off-line analysis of historical alarm data and for on-line alarm filtering.

Some problems of the current load balancing algorithms for distributed systems include:1) The role of each node in the system is fixed, and the system has no adaptability. 2) The load balancing algorithm is not universal. 3) The migration task is too large, and the load balance cycle is too long. To solve these problems, a hybrid load balancing algorithm was proposed. Firstly, a distributed receiving system model was designed, by which the system tasks were divided into three parts:receiving level, handling level and storing level. In receiving level, a home-made transmission protocol was used to improve the reception capability of the system. And then, in the load balancing algorithm, random load migration strategy was used. According to the status of the nodes, the tasks of load were randomly migrated. The problems of long load balance cycle and load moving back were solved by this strategy. Finally, the distributed control node selecting strategy was adopted to make the nodes adaptable. The experimental results show that the average delay in each layer of the system is in milliseconds, and the system load balancing takes less than 3 minutes, which proves that the load balancing mechanism has short load balance cycle and fast response, and can improve the reception capability of the distributed system.

The mining and analysis of smart traffic card data can provide strong support for urban traffic construction and urban management. However, most of the existing research data only include data about bus or subway, and mainly focus on macro-travel patterns. In view of this problem, taking a city traffic card data as the example, which contains the multi-source daily travel data of urban residents including bus, subway and taxi, the concept of tour chain was put forward to model the behavior of residents. On this basis, the periodic travel characteristics of different dimensions were given. Then a spatial periodic feature extraction method based on the longest common subsequence was proposed, and the travel rules of urban residents were analyzed by clustering analysis. Finally, the effectiveness of this method was verified by five evaluation indexes defined by the rules, and the clustering result was improved by 6.8% by applying the spatial periodic feature extraction method, which is helpful to discover the behavior pattern of residents.

Aiming at the problem that most of the existing time-domain modal parameter identification methods are difficult to set order and resist noise poorly, an unsupervised learning Convolution Neural Network (CNN) method for vibration signal modal identification was proposed. The proposed algorithm was improved on the basis of CNN. Firstly, the CNN applied to two-dimensional image processing was changed into the CNN to deal with one-dimensional signal. The input layer was changed into the vibration signal set of modal parameters to be extracted, and the intermediate layer was changed into several one-dimensional convolution layers, sampled layers, and output layer was the set of N-order modal parameters corresponding to the signal. Then, in the error evaluation, the network calculation result ( N-order modal parameter set) was reconstructed by the vibration signals. Finally, the squared sum of the difference between the reconstructed signal and the input signal was taken as the network learning error, which makes the network become an unsupervised learning network, and avoids the ordering problem of modal parameter extraction algorithm. The experimental results show that when the constructed CNN is applied to modal parameter extraction, compared with the Stochastic Subspace Identification (SSI) algorithm and its Local Linear Embedding (LLE) algorithm, the convolutional neural network identification accuracy is higher than that of the SSI algorithm and the LLE algorithm under noise interference. It has strong noise resistance and avoids the ordering problem.

Aiming at the problems of limited hiding capacity and obvious increasing bit rate in the existing hiding solutions, an intra-frame video information hiding algorithm based on diamond coding was proposed. Firstly, based on High Efficiency Video Coding (HEVC), two prediction models of adjacent 4×4 blocks were combined into a pattern pair, then the improved diamond coding algorithm was used to guide pattern modulation and information embedding. Next, the embedding coding for hidden informtion was done for second time with keeping the optimal coding division, thus ensuring the embedding quantity and eliminating intra frame distortion drift. The experimental results show that the Peak Signal-to-Noise Ratio (PSNR) is reduced by less than 0.03dB and the bit rate is increased by less than 0.53% by using the proposed algorithm, while the embedding capacity is greatly improved, and both the subjective and objective qualities of the video are well guaranteed.

The neural activity signal of interest is often influenced by structural noise and random noise in functional Magnetic Resonance Imaging (fMRI) data. In order to eliminate noise effects in the analysis of activate voxels, the time series of voxels preprocessed by Statistical Parametric Mapping (SPM) were transformed by Activelets wavelet. After getting scale coefficient and detail coefficient, the two kinds of noise denoised were eliminated separately according to their corresponding characteristics. Firstly, the Independent Component Analysis (ICA) was used to identify and eliminate the structural noise sources. Secondly, an improved algorithm for spatial correlation was presented on the detail coefficient. In particular, in the improved algorithm, the voxel similarity in the neighborhood was used to determine whether the detail coefficient reflected the noise or the neural activity. Experimental results show that the processing of data effectively eliminate the effect of noise; specifically, the frame displacement decreased by 1.5mm and the percentage of spikes decreased by 2%; in addition, the false activation regions are obviously restrained in the spatial map got by denoised signals.

In view of the problems that the spatial distribution of traffic model in traditional Network on Chip (NoC) was not consistent with the communication locality in practical applications and the overhead of network bandwidth is large, a novel algorithm for flow generation with NoC localized characteristic based on Rent rule was proposed. By establishing the communication probability distribution model with finite Mesh structure, the communication probability matrix was used to send packets to each node uniformly and obtain synthesis flows, and the locality was realized. The experiment simulated on flow with different locality degree and different network size. The results show that the proposed algorithm has better performance in flow locality, which is more close to the actual flows compared with five algorithms including Random Uniform, Bit Complement, Reversal, Transpose and Butterfly. In addition, the overhead of network bandwidth is lower.

In the traditional Fuzzy C-Means (FCM) clustering algorithm, the initial clustering center is uncertain and the number of clusters should be preset in advance which may lead to inaccurate results. The fuzzy clustering algorithm based on midpoint density function was put forward. Firstly, the stepwise regression thought was integrated as the initial clustering center selection method to avoid convergence from local circulation, and then the number of clusters was determined, finally according to the results, the validity index of fuzzy clustering including overlap degree and resolution was judged to determin the optimal number of clusters. The results prove that, compared with the traditional improved FCM, the proposed algorithm reduces the number of iterations and increases the average accuracy by 12%. The experimental results show that the proposed algorithm can reduce the processing time of clustering, and it is better than the comparison algorithm on the average accuracy and the clustering performance index.

The traditional Point-wise Mutual Information (PMI) method has shortcoming of overvaluing the co-occurrence of two low-frequency words. To get the proper value of k of improved PMI named PMI^k to overcome the shortcoming of PMI, and solve the problem that the term extraction cannot be obtained from a segmented corpus with segmentation errors, as well as maintaining the portability of term extraction system, combining with the PMI^k method and two fundamental rules, a new method was put forward to identity terms from an unsegmented corpus. Firstly, 2-gram extended seed was determined by computing the bonding strength of two adjoining words by PMI^k method. Secondly, whether the 2-gram extended seed could be extended to 3-gram was determined by respectively computing the bonding strength between the seed and the word in front of it and the word located behind it, and then getting multi-gram term candidates iteratively. Finally, the garbage of term candidates were filtered using the two fundamental rules to obtain terms. The theoretical analysis shows that PMI^kcan overcome the shortcoming of PMI when k≥3(k∈N₊). The experiments on 1 GB SINA finance Blog corpus and 300 MB Baidu Tieba corpus verify the theoretical analysis, and PMI^k outperforms PMI with good portability.

The existing error concealment algorithms for I frame are difficult to balance the recovery image quality and the algorithm complexity. To solve the problem, an efficient intra-fame partitioning error concealment method was proposed. Firstly, according to the motion correlation between video frames, the lost macro blocks were divided into motion blocks and static blocks. For static blocks, frame copy error concealment method was used to conceal lost blocks. For motion blocks, they were divided into smooth blocks and texture blocks by the texture information of the correctly decoded macro blocks. Then, the bilinear interpolation method was adopted to restore the smooth blocks and more delicate Weighted Template matching with Exponentially distributed weights (WTE) method was used to conceal texture blocks. The experimental results show that, compared with the WTE method, the proposed method has improved the Peak Signal-to-Noise Ratio (PSNR) by the average of 2.6 dB and decreased the computation complexity averagely by 90%. As for video sequences with different features and resolutions in continuous scene, the proposed method achieves certain applicability.

The single frequency modulation or phase modulation waveform based on chaotic sequence has low waveform complexity, which limits predictive probability of chaotic signal, radar intercept probability and anti-interference performance. In order to solve the problems, joint frequency-phase modulation based on chaotic sequence in radar waveform was proposed. Firstly, the radar signal was carried out for the chaotic frequency encoding, which was that a pulse was divided into a series of sub-pulses and different frequency modulation was carried out for different sub-pulses. At the same time, in each frequency encoding sub-pulse, the random initial phase was used in each cycle of waveform. The simulation results show that the maximum value of autocorrelation sidelobe peak of joint frequency-phase modulation based on chaotic radar signal achieved -24.71 dB. Compared with the frequency modulation or phase modulation based on chaotic signal, the correlation performance of the proposed joint frequency-phase modulation has improved. The experimental results show that, the joint frequency-phase modulation chaotic radar waveform combines the advantages of phase modulation and frequency modulation and is an ideal detection signal with the flat power spectrum characteristic of phase modulation and anti-noise-interference ability of frequency modulation.

A smart and green home is a dynamic large-scale system with high complexity and a huge amount of information. In order to further improve coordination between subsystems and make the best of multi-source information for the smart home, a multi-Agent intelligent home system based on multi-source information fusion was designed. The framework and interaction mechanisms of Agent were introduced and a multi-source information fusion model based on Adaptive Neural-network-based Fuzzy Interference System (ANFIS) was put forward to conduct the feature extraction and learn occupant's personal behavior. A simulation platform using lightweight embedded Jade Agent on Android and Matlab on personal computer was developed to control the natural lighting system in smart home. The theoretical analysis and the simulation results show that the model can improve synergistic interaction of home systems, and finally enhance the efficiency of multi-source information fusion in decision making process.

In order to review all the interested contents from large volume recorded surveillance videos in very short time, a video digest method based on target indexing was presented in the paper. On the basis of optical flow analysis, the background got updated in the still region and the moving object image got segmented from the moving area. Accelerated motion model constrained the candidate points for corresponding feature tracking. Successful establishment of trajectory became the basic reference to cluster the whole objects into several classes. All the objects parameters such as size, velocity, and color were saved in XML format as the indexes. Finally, the living digest video was reproduced by mosaicking the selected objects images on the background frame by frame in the same order as original. Since the method removes all the spatial and temporal redundancy, a much shorter living video could browse all useful information of a long surveillance one.

Concerning the large volume of data of multi-view video transmission system, an Unequal Error Protection (UEP) method based on depth-map and Distributed Video Coding (DVC) was proposed. Firstly, the depth-map based on multi-view was extracted. Secondly, one point view and the depth-map were transmitted. Finally, through the network transmission, other point views were generated by the point view and the depth-map at the decoder. Considering their different importance at the decoder, different distributed video coding methods were used in the point view and the depth-map to realize UEP. The simulation results show that the proposed transmission method provides stronger error resilience and higher transmission reliability, and it improves the Peak Signal-to-Noise Ratio (PSNR) of images about 1. 5 dB than traditional distributed multi-view video coding transmission system.

It is essential for software development and maintenance to predict which modules are change-prone in an Object-Oriented (OO) software system. In this paper, a light-weight approach was developed to compute the change probability metric by leveraging the dependence relationships between classes in a system. Then, based on Logistic regression model, an experimental analysis was conducted using Eclipse 2.0. The experimental results indicate that, on one hand, the proposed change probability metric captures different information from traditional OO metrics. On the other hand, when being used with traditional OO metrics together, the proposed change probability metric can significantly improve the accuracy for predicting the change-prone classes.

In order to improve the face detection test results under the influence of illumination change and complex background, an algorithm of 3D color clustering model based on direct least squares estimate was proposed during the preprocessing phrase. Firstly, three plane projection distributions of skin color were seen as fitting objects in CbCrCg space, and then smooth edge was got by median filter and Sobel operator, at last the best 3D color model was got through direct least squares. In experiment, the public face library and face image got by outdoor shooting were seen as objects, and the experimental results show that, this algorithm has better segmentation effects than traditional color preprocessing algorithm, and it has improved the detection rate more effectively.

Because the base stations can only be placed at the border of the monitored area, the border placement problem was formally defined. For the goal to place the minimum number of base stations to cover as much as possible the monitored areas, an improved placement algorithm with polynomial time was proposed. The coverage percentage of initial algorithm was analyzed first. When initial coverage percentage is larger than guaranteed coverage percentage, it is possible to reduce the size of initial placement set. Finally, placement set was gradually improved to achieve the minimun of placement set. The results indicate that the coverage percentage and placement set of the proposed algorithm are superior to random algorithm in different test environments.

Coverage problem is one of the most fundamental problems in wireless sensor networks since it reflects the sensing quality. The present studys mostly concentrates in Omni-directional sensors which is not suitable in many applications such as video surveillance systems consisting of directional video sensors. This paper present a new (k,ω)-angle coverage problem which study directional sensors deployment. The goal is to deploy minimal number of sensors to k-angle cover all the targets. It present a greedy algorithm to solve this problem. For this algorithm, it define three contribution functions to determine the location to deploy sensor. The proposed method greedily selects a maximal contribution location to deploy a sensor until the entire targets are k-angle covered. Simulation results exhibit the characteristic and performance of this algorithm.