Abstract:Aiming at the problems of high cost and timeconsuming in volumetric error detection and identification of multiaxis machine tools, a novel identification method for linear axis positiondependent geometric error based on double ballbar test is proposed. Firstly, the error models of the axis motion in various planes are established, respectively; and the error element of each axis is prefitted with respect to its position using polynomial. Then, the comprehensive error in each plane is obtained with conventional threeplane circular arc trajectory measurement. Finally, the fitting coefficients are solved based on the least square method, which replaces the traditional method that directly solves the specific value of error element; and the positiondependent geometric error element identification of linear axes is realized. The experiment results verify the correctness and efficiency of the proposed method. The method has reference significance for the linear axis error identification and compensation of CNC machine tools.

Abstract:Anovelhighprecision time interval measurement method is proposed in this paper. The short pulse is utilized to representthe event to trigger a specially designed highspeed ring oscillator circuit. In this way,the clock signal is generated, which issynchronized with the event.It is subsequently adoptedas the sampling clock of ADC to sample thesinusoidal reference signal. Accordingly, the time interval between the two events is mapped to the initial phase difference between the two points on the sinusoidal reference signal. The allphase fast Fourier transform (apFFT) algorithm is performed on a finite number of samples to accurately determine the initial phase difference.Then,the accurate time interval between the two events is achieved. The proposed method reduces the difficulty of engineering implementation and increases the reliability and practicability of the measurement. In the case of a sinusoidal reference signal with 10 MHz frequency, a 12bit ADC with 133 MHz sampling frequency, and apFFT with operation point number N=4 096, the singleshot time interval measurement precision of 28 ps rms and time resolution of about 1 ps can be realized. The error distribution is close to normal distribution. Experimental results are in agreement with the error budget of the theoretical analysis.

Abstract:On the base of hardware performance of spectrometer, wavelength calibration is the key process of further introducing errors. If the error level introduced in the solidification process of the calibration equation coefficients equals the pixel sampling error, then the wavelength measurement error in practical application includes two errors that comprises the pixel sampling error and the error introduced in the wavelength calibration equation calculation, which will enlarge and double the measurement error. Using the independently developed compact spectrometer, this paper aims to study the influence factors in wavelength calibration through data analysis, and then give an effective method to avoid error expanding. The FHWM of the adopted spectrometer is about 5 nm for a 100 μm slit, the wavelength interval of the pixels is about 045 nm, and the wavelength error of the hardware system is theoretically a halfpixel wavelength interval of 0225 nm. The influence of the measurement noise and peak finding algorithm on calibration accuracy is analyzed emphatically, and the averaging noise reduction with multimeasurements combined with Gauss fitting peak finding method is proposed to improve calibration accuracy. Compared with traditional direct extremum method, the 2 times standard deviationof the fitting residual of calibration equation for the proposed method is about 01 nm, while the 2 times standard deviation of the fitting residual of calibration equation for traditional direct extremum method is 037 nm (two methods both use the 5th order polynomial as the calibration equation). Study reveals that the three key factors of error control in wavelength calibration are measurement noise control, peak finding algorithm and least squares adjustment. Through selecting appropriate algorithm parameters, the standard deviation of the fitting residual of wavelength calibration can be controlled at the level of about 1/10 pixel wavelength interval, which fully reflects the multipixel statistical advantage of Gauss fitting peak finding algorithm.

Abstract:Intelligent swarm based on unmanned aerial vehicles or robots has become a research hotspot. To achieve collaborative control of dense cluster, the key information is to achieve the precise relative distance among adjacent agent clusters. Ultrawideband (UWB) wireless transmission technology can transmit information and realize centimeterlevel theoretical ranging accuracy. It has broad application prospect in coordinated navigation and control of clusters. In this paper, the UWB ranging scheme for centimeterlevel ranging requirement of relative navigation is first described. Then, the reason and characteristics of the ranging error caused by the clock offset of UWB antenna in the actual environment, the relative speed between the nodes and the nonlineofsight environment is analyzed. The state detection, error compensation and estimation methods of UWB distance measurement are studied. Finally, based on the DW1000 ultrawideband module, the experimental environment is constructed and the method studied in the paper is evaluated. Experimental results show that the error estimation and elimination method can significantly improve the accuracy of UWB in practical application. Compared with the traditional ranging algorithm, the ranging accuracy of the proposed method is improved under the relatively static scenario and a scene with relative speed between nodes. To be specific, the error is reduced by 70% in the lineofsight environment state and by 50% in the nonlineofsight environment state.

Abstract:Antenna measurement is very important for the performance verification of antenna radiation. However, when the antenna size is relatively large or the test frequency is relatively low, the traditional measurement method cannot meet the test requirement. In this paper, one kind of antenna measurement method based on the micro UAV is investigated for testing the large antenna in the outdoor environment. The gain pattern of measured antenna is obtained according to the measurement channel model. In this method, the realtime differential GPS is used as the navigation signal, and the geometric control method is combined to control the exact flight trajectory of UAV more accurately. Compared with the traditional navigation method based on GPS, the flight trajectory deviation of UAV is less than 01 meter. Experimental results show that the method can measure the antenna pattern precisely. The gain measurement error is less than 1 dB, which has high engineering application value.

Abstract:Existing absolute measurement methods take a lot of time to align and adjust the tested flat precisely during the measurement process to ensure the high accuracy. In this paper, we propose a twoflat measurement method that does not require high precision adjustment of the tested flat. The method requires the tested flat measured three times, including the original position, the unknown angle after rotating, and the unknown distance after lateral shifting. The actual rotation angle and shift displacement of the tested flat are solved by feature matching. The surface of the tested flat is reconstructed by the iterative algorithm and Zernike polynomial fitting. Meanwhile, the compensation of the lost angular frequency terms is considered. This method avoids the precise adjustment and shorten the measurement time while ensuring the measurement accuracy. Compared with Vannoni′s method, experimental results show that root mean square (RMS) of the residual of the two measurement methods is 0004λ, and the adjustment process of the tested flat is fast, simple and coherent. The proposed method provides a new way of surface measurement for optical components.

Abstract:Aiming at the issue of fault acoustic emission (AE) signal intelligent recognition of rolling bearing under multiple working conditions, a new fault recognition method combining long shortterm memory (LSTM) networks and transfer learning (TL) is proposed. This method only takes the original AE signal parameters under single working condition as the training samples and constructs LSTM model to fully excavate the deep mapping relationship between AE signals and faults, so as to identify the faults under other working conditions that have similar distribution characteristics with the training working condition. TL is introduced and combined with the LSTM model to deal with the fault identification problem under other working conditions that have different distribution characteristics. Thus, the adaptive extraction and intelligent recognition of the fault features under various types of working conditions can be completed. The experiment results show that the recognition of inner ring, outer ring and cage faults have high accuracy under the working condition changes of the rotation speed, acquisition position and type of the rolling bearing. The realtime online intelligent monitoring task of the faults can be completed endtoend under various types of working conditions. The proposed method gets rid of the overreliance on prior fault data, and the feasibility and superiority of the proposed method are verified.

Abstract:In remote field eddy current testing(RFECT)of ferromagnetic pipes, because the eddy current signal penetrates the pipe wall twice from the exciting winding and pickup winding, the same defec twill affect the propagation of the remote field eddy current signal twice,which leads to the appearance of two peaks in the received signal, they are called real peak and pseudopeak here. The real peak contains the useful information such as the position of the pipe defect, whereas the pseudopeak acts as a disturbance to the defect positioning in the pipeline test and evaluation. In order to eliminate the disturbance of the pseudopeak to defect positioning, this paper proposes a novel pseudopeak removing method of remote field eddy current. Firstly, mathematical morphological filter(MMF) is utilized to eliminate the baseline bias of the defect signal, then a mathematical model based on radial basis function(RBF) is introduced to fit the real peak and pseudopeak, and the NelderMead simplex method(NMSM) is used to solve the model parameters and obtain the pseudopeak signal. Simulation experiment and practical data processing results indicate that the proposed method possesses a decent pseudopeak removing performance and certain applicability.

Abstract:Aiming at the development trend of microgrid energy management technology at the present stage, on the basis of satisfying its internal economic dispatching, attention should also be paid to the energy complementation mechanism among microgrids. In this paper, a bilevel energy optimal dispatching model for gridconnected regional microgrid cluster is proposed. Conditional value at risk (CVaR) is introduced to measure the impact of renewable energy sources and load forecasting errors on dispatching scheme, which is taken as the optimization objective of internal energy dispatching in the microgrid combining with the microgird operation income.The multiobjective particle swarm optimization (MOPSO) algorithm is adopted to obtain the solution. The incomerisk ratio is formulated as the screening index of optimal dispatching strategy, and the internal energy optimal dispatching strategyin the microgridis proposed. On the premise of minimizing the active power gradient variation at the common coupling point of the regional networked microgrid cluster, the optimal combination scheme of thenet power of the microgrids is obtained, which can suppress the power fluctuation caused by the microgrid cluster to distribution network. Then, considering the power transmission distance, the net power complementation mechanism among the microgrids is formulated to improve power transmission efficiency. The simulation results of examples show that the model can reasonably realize the economic operation and power balance within the microgrids and amongthe microgrids, and provide effective design process for the dayahead dispatching plan of the microgrid cluster.

Abstract:Empirical Modal Decomposition (EMD) and the methods based on EMD have been widely used in the field of fault diagnosis. The selection of Intrinsic Mode Function (IMF) after decomposition is important for accurate extraction of fault features. To solve such problem more effectively, the gearbox local fault optimal feature extraction algorithm based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) combined with Signal Quality Index (SQI) algorithm and Singular Value Decomposition (SVD) is proposed in this study. The method is evaluated by experiments on local crack of gear with different fault levels. Firstly, the original data are obtained by experiment. Then, they are decomposed by CEEMDAN. The effective IMF is decomposed by SVD to obtain the optimal feature vector, which is the input of BP neural network for training and test. Finally, the test results are compared with several common methods. Experimental results show that the proposed CEEMDANSQISVD algorithm has high recognition accuracy and is better than several conventional methods for local fault of gearbox.

Abstract:Aiming at the error accumulation problem ofthe visualinertial odometry algorithm based on multistate constraint kalman filter (MSCKF) in the augmentation process of the camera state equation, a multimode augmentation method of camera state equation is proposed. In this method, the stability of the visual feature tracking state is strictly judged firstly; then, two methods are automatically selected to augment the camera state equation, the first method optimally solves the relative pose parameters of the camera based on visual image information,another method is based on the recursion results of inertial measurement unit (IMU) state combining the cameraIMU external parameters to initialize the camera pose parameters for new image frame. As a result, the error accumulation problem of IMU under the stable feature tracking state is solved. In the experiment part, the performance of the proposed algorithm is verified utilizing the EuRoC dataset and practical application dataset. The experiment results show that the improved MSCKF algorithm can effectively avoid the error accumulation of IMU under the stable feature tracking state,further fuse the complementary advantages of both visual and inertial systems,and improve the localization & orientation precision and stability of the carrier.

Abstract:There is a limitation existing in classical bearing fault diagnosis thatit is required to build different target models to fitvarying working conditions. This paper proposes a spectral centroid transfer learning model, which transfers the source working condition domainto target working condition domain; the modeling cost for target working condition domainis reduced and the universality of bearing fault diagnosis modelis enhanced.Firstly, the frequency spectrum similarity measure (FSSM) value between the two working conditions domains is calculated and the source working condition domain with near distance is sorted and selected as the initial training set. Then, during iteration process, the samples whose spectral centroid meandistances are relatively far from that of the training set are removed, and the same quantity of label less samples from target working condition domain are added to training set. The iteration finishes when the spectral centroidme and istances of both the working conditions domainsare equal. Here the fault categories are determined by the outputs of two subclassifiers: The support vector machine (SVM) and the logistic regression (LR)based subclassifiers. The experiment results on Spectra Quest geared drive train show that the diagnostic performance of the proposed model is significantly better than that of nontransfer model when the rotation speed or load changes. Meanwhile, some indexes, including the number of replaced samples, the diagnostic accuracy, the FSSM index and the time consumption can be utilized to evaluate the quality of the source working condition domain.Thus,the proposed model possesses a potential valuein solving bearing fault diagnosis issue under varying working conditions.

Abstract:The filter hardware structure design of traditional back EMF method will lead to the commutation error of the position sensorless brushless DC motor. When the commutation error is large, the performance index of the system will be greatly reduced. Aiming at the problems of system commutation torque ripple and outofstep caused by commutation error, the relationship between the ideal backEMF zerocrossing point and the actual backEMF zerocrossing point is analyzed; on the base of adding a singleended primary inductor converter (SEPIC) predriver circuit, this paper adopts PWMONPWM modulation method, which detects and compares the voltage value of the nonconducting phase (i.e., the suspended phase) and the calibrated voltage value, then a commutation signal is obtained and the commutation error correction is realized. Compared with the traditional backEMF method, the proposed method needs only to detect one phase voltage, then the commutation error is accurately corrected and the torque ripple of the system is reduced at the same time. Simulation and experiment results verify the effectiveness of the proposed commutation error correction strategy in full speed range.

Abstract:he stability and reliability of the measurement tube vibration is the basis for Coriolis mass flowmeter to achieve accurate measurement. Aiming at the vibration amplitude control issue of Coriolis mass flowmeter, a humansimulated intelligent control method is designed. The vibration of Coriolis mass flowmeter measurement tube is divided into two stages of vibrationstart amplitude stabilization and interference suppression; according to the amplitude deviation and deviation rate, five and three different characteristic states are decided, and the characteristic state sets are constructed. The corresponding control modes are designed for different characteristic states respectively, and the control mode sets are constructed, and the heuristic search and intuitive inference rule sets are designed. In the control process, the current vibration state is identified according to the deviation and deviation rate, and corresponding control mode is selected according to the inference rule set to conduct the vibration control. The experiment results show that the proposed method can achieve fast oscillation starting and stable amplitude of the Coriolis mass flowmeter, which verifies the superiority and engineering practicability of the proposed method.

Abstract:Electromechanical actuator (EMA) has become the core components of more/all electric aircraft due to its light weight, small size, high reliability and etc., which has gradually been widely utilized in various types of more/all electric aircrafts. However, the dynamic operation profiles and load conditions of EMA bring great challenges to its degradation modeling and health indicator (HI) estimation. Therefore, an EMA HI construction method based on Bayesian updating is proposed. Firstly, an HI prior model is built based on historical monitoring data. Secondly, the EMA HI prior model parameters are updated iteratively through utilizing Bayesian updating theory and realtime monitoring data. Finally, the EMA degradation state under various operation profiles and load conditions is accurately characterized. This study provides a novel idea for solving the mismatch issue of the EMA HI construction model under variable working conditions, and experiments with NASA benchmark data verify the effectiveness of the proposed method. The experiment results show that compared with the EMA HI construction method based on model identification, the proposed EMA HI construction method based on Bayesian updating has stronger adaptability to the working conditions and can effectively construct EMA HI under variable working conditions.

Abstract:Studying the influence of the strain gauge layout on the output of the force transducer under eccentric and tilting loading is very important to enhance the antieccentric and tilting loading ability of the transducer. In this paper, using the theory of elasticity, the output model of the column force transducer under eccentric and tilting loading is established considering the influence of the position deviation, angle misalignment and height location of the strain gauges. Based on the model, the influences of the position deviation, angle misalignment and height location on the indication error and direction error are discussed, respectively. The verification experiments were carried out. Study results show that under eccentric and tilting loading, when position errors exist among the strain gauges or angle misalignment of the strain gauges exists, large indication error and direction error occur. The indication error and direction error increase sharply as the position deviation and angle misalignment increase. When attaching the strain gauges, if the position deviation and angle misalignment are both not greater than 1 degree, the indication error could be controlled within ±015% even the transducer is under the tilting loading of 01 degree. The height location variation of the strain gauge has a little effect on the output of the transducer. The analysis results of the developed force transducer output model are accurate and reliable, which can be a guidance to control the manufacture processing of the force transducers.

Abstract:metal doping; gas sensor array; gas detection; temperature modulation; multioutput support vector regression

Abstract:plastic deformation; nondestructive testing; electromagnetic acoustic resonance

Abstract:As a branch of haptic interaction technology, wearable haptic interaction technology can provide operators with a more realistic and natural humancomputer interaction experience. In this work, a wearable fingertip haptic interaction device is designed. The device consists of a control box worn on the wrist and a structure worn at the fingertip, and the structure weighs only about 30g with a size of 38 mm×31 mm×50 mm. The device uses linkage mechanism and can realize 3 DOF haptic feedback. The device adopts bluetooth communication and is small, lightweight and easy to wear. The direction perception and recognition experiment indicate that the device can produce 8 effective directional information on the finger pad, and each direction can be correctly perceived by subjects. In addition, the virtual environment interaction experiment proves that the device can help people control the interaction force in the virtual environment and increase the efficiency of humancomputer interaction.

Abstract:Traditional lower limb prosthesis motion intent recognition methods often use multimodal sensor signals, which bring certain complexity and lags of pattern transition recognition. This paper proposes a datadriven based intelligent lower limb prosthesis motion intent recognition method. After redefining the movement patterns of unilateral lower limb amputees, only the inertial sensorsare used to collect the time series data in the swing phases of the healthy side. The Gaussian mixture modelhidden Markov model (GMMHMM) is selected as the classifier to recognize the motion intent of lower limb prosthesis. The experiment results show that the recognition rate of the method reaches 9899% in steady patterns: levelground walking, ramp ascent, ramp descent, stair ascent and stair descent, and 9692% in 13 motion patterns that contain 5 steady patterns and 8 transition patterns. The method proposed in this paper can greatly improve the recognition performance of lower limb prosthesis motion intent, and help the unilateral lower limb amputees to walk naturally, smoothly and steadily.

Abstract:Aiming at the key issue of trajectory tracking control for domestic laparoscopic surgery robot, an adaptive fuzzy sliding mode control system is constructed based on dynamic model. This method uses a fast nonsingular terminal sliding mode surface to ensure that the system can converge in a limited time and avoid the singularity of control. At the same time, the fuzzy logic control is used to dynamically control the size of the sliding mode switching term according to the dynamic variation of the system error. The chattering of the system is reduced, and the tracking accuracy and robustness of the controlled system are improved on the basis of retaining the robustness of sliding mode control. The simulation results show that the designed method possesses the characteristics of fast convergence speed, high steadystate accuracy and small torque ripple. The method can reduce the position convergence time from 06 s～09 s to 01 s and the peak value of the steadystate error reduces from 0035 rad to 001 rad. The experiment results on living animal tissues show that the designed control system can achieve the aim the mechanical operating arm accurately tracks the operation information of the doctor, which provides a guarantee for the smooth implementation of the operation.

Abstract:Pianoplaying glove is one kind of emerging intelligent wearable equipment. By using the multiinertial sensors in glove, the gesture of piano player can be realtime perceived and analyzed. The learners can know in realtime whether the playing gesture is right.Thus,the efficiency and interest of piano learning can be improved and the cost of learning can be reduced effectively. Different from gestures in other application fields, piano playing gestures have the characteristics of diversity, rapidity, large dynamics and strong timevarying. In this study, the piano playing gesture recognition system based on inertial data glove and infrared detecting rod is designed. A method of gesture recognition for piano playing based on machine learning is proposed. The output of inertia data gloves and infrared detection rods are used as data sample. According to the characteristics of piano playing gestures, multimodal gesture features are extracted.Hierarchical recognition algorithm is adopted to improve the recognition effectiveness. Experimental results show that the proposed recognition method can better meet the needs of gesture recognition in piano playing. The recognition accuracy rate is better than 99%.

Abstract:Aiming at the feature extraction issue of four class motor imagerytask,this paper proposes an EEG signal feature extraction method based on intrinsic timescale decomposition (ITD) and phase synchronization analysis.The fourclass motorimagery datasets from the BCI Competition III and BCI Competition IV are adopted.Firstly,this method selects five channel motorimagery EEG (electroencephalogram) signals,calculates the phase locking value(PLV) among the channels according to the phase synchronization,and uses the PLV as a kind of feature. Then,ITD is used to decompose the five channel motorimagery EEG signals and extract the energy feature of the first layer proper rotation component(PRC), which is combined with the PLV feature to obtain the fifteendimensional feature vector. Finally, support vector machine (SVM) is used for classification recognition. The average recognition rate and Kappa coefficient for 12 subjects reach 9164% and 0887,respectively. The results show that this method can effectively extract the feature of EEG signals and improve the classification accuracy of fourclass motor imagery task.

Abstract:The traditional evaluation method of cardiac chest compression is based on acceleration waveform integration, which is affected by noise and integral delay. A large error is introduced in calculating the distance and the evaluation effectiveness is not ideal. Therefore, based on the weak supervised learning strategy and waveform segmentation, this paper proposes one kind of acceleration waveform recognition algorithm for cardiac chest compression based on onedimensional convolutional neural network. Experimental results show that the onedimensional convolutional neural network achieves 994% accuracy, which is significantly better than the traditional integration method and BP neural network algorithm. Further, the GradCAM method is adopted to visually analyze the evaluation results. The features of convolutional neural network focus on the acceleration waveform changes in the two compression stages of starting to press until the pressure can reach the equilibrium position. The reverse acceleration can achieve the maximum value after the pressing of the hand to the next press start. In addition, the evaluation model does not need to accurately measure the pressing distance. Thus, it is not affected by factors such as pressing occlusion and electromagnetic wave interference. Its effectiveness can be checked in the realtime manner. It also has the feature of high robustness in complex environment and has certain practical value in the field of medical emergency.

Abstract:The current somatosensory selective attention based brain computer in terface (BCI) system has disadvantage of less command for multidegree deviceandlow information transmission rate. To solve these problems,a novel hybrid BCI system combing steadystate visual evoked potential (SSVEP) and somatosensory selective attention (SSA) is proposed in the paper. The SSVEP and event related desynchronization (ERD) can be elicited withtheaidof visual and somatosensory stimuli. In order to overcome the shortcomings of conventional feature extraction method which needs more heuristic knowledge, a deep learning algorithm is used to decode the EEG signal. In this method, the temporaldomainsignals of several channels are converted into temporalfrequencyspatial domain feature image. Eight subjects arerecruited to participate the experiment. The average accuracy of offline test is 8135%, which indicates that the proposed multimodal hybrid BCI based on SSVEP_SSA is feasible for instruction set extension and decoding precisely.

Abstract:In order to meet the requirement of high accuracy and strong antinoise performance of image edge localization in computer vision calibration and precision measurement, a subpixel level image edge detection algorithm based on Franklin moments is proposed. Firstly, a subpixel edge model is established to extract the detailed features of image edge points with the convolution of Franklin moments at all levels. Then, according to the rotation invariance principle of Franklin moments, the relationship among different levels of Franklin moments after image edge rotating to the vertical direction is analyzed, and the key parameters of the subpixel image edge are determined. Finally, the actual subpixel edge points in the image are located based on the improved edge judgment condition. A large number of experimental results show that compared with Zernike moment subpixel level image edge detection algorithm, the subpixel level image edge detection algorithm combining wavelet transform with Zernike moment, and the subpixel level image edge detection algorithm combining Roberts operator with Zernike moment, the proposed subpixel level image edge detection algorithm based on Franklin moments possesses higher speed and accuracy and stronger antinoise performance, which better meets the measurement requirements of stability and reliability and high precision in image edge localization.

Abstract:Aiming at the defect that existing automatic reading algorithm of the pointer meter has strict requirements on the image acquisition conditions. This paper proposes an automatic reading method of the pointer meter based on double Hough space voting.According to the distribution characteristics of the meter scale and the center of the circle in Hough space, the method adaptively calculates the center of the meter, divides the image in the polar coordinate space with the projection method, extracts the scale and pointer information, and finally reads the pointer meter with the distance method. Experiments show that the quoted error of the proposed algorithm is below 08%, which effectively improves the correct rate and robustness of the meter reading.

Abstract:Autonomous mapping for mobile robot is the premise of completing intelligent behavior.To improve the intelligence and intuitive user interaction of robot, maps are needed to achieve the semantics beyond geometry and appearance. This paper studies the 3D semantic map construction method, which fuses the pixellevel image semantic segmentation based on Deep Residual Networks(DRN) and Simultaneous Localization And Mapping(SLAM).Firstly, the combined median filter algorithmis used to restore the depth of the map. The improved Iterator Closest Point (ICP) algorithm is employed to estimate camera pose and loopback detection based on random ferns is proposed for 3D scene reconstruction. Then, the optimized DRN is utilized to achieve more accurate semantic prediction and segmentation. Finally, the predicted semantic classification labels are migrated to the 3D model by Bayesian based incremental transfer strategy to generate a globally consistent 3D semantic map.Experimental results show that the proposed method can build the realtime 3D semantic map in the real and complicated environment.

Abstract:Aiming at the sceneswhere contact rotation angle measurement devices such as decoding device are not applicable, apure attitude measurement algorithm based on monocular visual odometer is designed, which possesses the advantages of low cost, good precision and no drift. In order to solve the application problem in pure rotation condition,based on the classical monocular visual odometer algorithm,adopting the methods of assuming depth, direct initialization, modifying attitude calculation and optimizing algorithm, the problem that the camera cannot workin pure rotation condition is solved. Aiming at the practical problem that the camera cannot be accurately installed on the rotation center of the measured object, an installation bias model is established and the attitude measurement algorithm is improved, which effectively solve the installation bias problem. Meanwhile, an offline measurement schemeis proposed,with which the installation bias parameter identification is achieved. The effectiveness of the algorithm was verified with a highprecision twoaxis turntable. The experiment results show that the proposed algorithm still possesses good comprehensive performance even using lowcost cameras. The multidegreeoffreedom dynamic angle measurement error is within 05 degree.