A novel binocular Visual Odometry(VO) algorithm is proposed for real-time precise localization of mobile robots.Firstly,it uses accelerated Scale Invariant Feature Transform(SIFT) operator to extract the image features on the left and right image.The sparse stereo matching is carried out after the extracting.In addition,the method of features tracking is applied between the previous and current image.Thus,the initial pose including rotation and translation matrix can be obtained with the motion estimation method based on the RANSAC strategy.Secondly,the image sequence is divided into key frames and non-key frames.In order to decrease the error of the inter-frame motion estimation,a variable sliding window is applied to optimizing the pose of adjacent key frames locally and nonlinearly.Finally,the closed-loop detection is applied by the method of bag of words.Furthermore,all the poses of key frames in the closed-loop are optimized globally to avoid the error accumulation and the drift of the trajectory.Experimental results show that the proposed algorithm has good real-time performance,while reducing the position pose error and improving the positioning accuracy.
The stability control strategy of biped robot walking process is studied,and a crus vibration control system based on the vibration acceleration of the upper body of the robot as feedback is designed.Through virtual simulation analysis software——Automatic Dynamic Analysis of Mechanical System(ADAMS),the virtual prototype of the biped robot is built and imported into Matlab.The automatic anti-disturbance control algorithm is designed to suppress the low frequency and high frequency according to active and passive vibration attenuation.Simulation results show that active and passive vibration attenuation greatly reduces the vibration of the upper body of the robot and compensates the situation that passive vibration can not suppress the low frequency vibration,and effectively improves the stability of the robot.
The function identification of Public Bicycle System(PBS) plays an important role for the balancing scheduling and layout planning of the system.As a new public transport mode,PBS has accumulated more and more data,which not only can reflect the social and economic activities of users at different times and locations,but also has a close connection with the function of the rental points.By combining the spate-time attributes of the leased point and through the historical data analysis of the system,this paper constructs a rental points clustering model of public bicycle system.It makes use of the Latent Dirichlet Allocation(LDA) model and K-means clustering algorithm to find the functional areas of the system,and through the analysis of the clusters’ use features and uses Point of Interest(POI) data and station names to verify the results.The results show that the proposed model can help the system managers grasp the function distribution of the rental points in PBS.
Before using D-S(Dempster-Shafer) theory for evidence fusion,the extent of the conflict among evidences should be determined.Aiming at this problem,this paper proposes a conflict measure and correction method based on trust degree and false degree.It calculates both trust degree and false degree of each evidence and sorts them respectively.The evidence with smaller trust degree and larger false degree is called high conflict evidence and done weighted correction.Then the Dempster combination rule is used for evidence fusion.Experiment results show that the proposed method can correctly judge the high conflict evidence and solve the problem that the fusion result contradicts the fact,while improving the speed and precision of convergence.
Traditional leader based and decentralized algorithm for solving Byzantine consensus exists the problem of low fault tolerance and high message complexity during the process of solving legitimacy verification.In order to solve these problems,this paper proposes a new blockchain consistency algorithm.It introduces two-phrase commit and quorum voting,using the characteristics of distributed ledge in Blockchain protocol to solve legitimacy verification and ultimate consistency is proved later on.Experimental result shows that compared with traditional Byzantine consensus protocol,this algorithm reduces the complexity of message passing and improves the system fault tolerance rate.
In order to segment interested objects in image accurately and quickly,an improved Wolf Pack Algorithm(WPA) is proposed combined with 2D maximum entropy to realize the effective segmentation of image targets.In travel link of WPA,the chaotic systems is used dynamically to adjust the inertia weight.In the attack link,chaotic global search is finisned in the whole solution space.The criterion function is maximized by using the improved WPA combining the 2D maximum entropy.Experimental results show that the proposed algorithm can get accurate image segmentation effect.Compared with the 2D maximum entropy combining with the origin WPA,it can improve the speed and accuracy of image segmentation.
In order to improve the precision rate of video smoke detection for wildland-urban interface,a video smoke detection algorithm based on multiple image features and deep learning is proposed.Firstly,ViBe(Visual Background extrator)method is applied to extract the moving or changing foreground areas in surveillance videos and image corner information is used to exclude the disturb from objects with detail texture.Secondly,color space features are utilized to narrow the check area for smoke.Thirdly,the cumulated difference image is calculated to find and eliminate the influence of rigid bodies.At last,the deep learning model is used to recognize the filtered image area.The algorithm is designed in the framework of cascade classifier and implemented by using parallel computing technology.Experimental results and project cases verify that the proposed algorithm is significant for improving the precision rate of smoke detection for wildland-urban interface.
The state detection method of eye and mouth is the key issue for fatigue detection,but it is affected by changing of illumination and wearing glasses.To solve above problems,a fatigue detection method based on facial behavior analysis is proposed.It designs an infrared video acquisition system for driver.The driver’s face is detected based on AdaBoost and the Kernelized Correlation Filter(KCF) tracking algorithm.The feature points are determined by the method of cascade regression,and the eye and mouth regions are obtained.Convolution Neural Network(CNN) is utilized to detect the state of eye and mouth.On this basis,the fatigue parameters are calculated for fatigue detection.Experimental results show that the method can detect the state of eye and mouth accurately and detect fatigue more effectively in many circumstances.