基于计算机视觉的前照灯光束照射方向检测方法

本文介绍了一种基于计算机视觉的汽车前照灯光束照射方向检测方法,采用面阵CCD 摄像机作为视觉传感器。可对远、近光进行检测。论述了系统结构和检测方法,并给出了部分图像处理结果。

Neighborhood fusion-based hierarchical parallel feature pyramid network for object detection

In order to improve the detection accuracy of small objects,a neighborhood fusion-based hierarchical parallel feature pyramid network(NFPN)is proposed.Unlike the layer-by-layer structure adopted in the feature pyramid network(FPN)and deconvolutional single shot detector(DSSD),where the bottom layer of the feature pyramid network relies on the top layer,NFPN builds the feature pyramid network with no connections between the upper and lower layers.That is,it only fuses shallow features on similar scales.NFPN is highly portable and can be embedded in many models to further boost performance.Extensive experiments on PASCAL VOC 2007,2012,and COCO datasets demonstrate that the NFPN-based SSD without intricate tricks can exceed the DSSD model in terms of detection accuracy and inference speed,especially for small objects,e.g.,4%to 5%higher mAP(mean average precision)than SSD,and 2%to 3%higher mAP than DSSD.On VOC 2007 test set,the NFPN-based SSD with 300×300 input reaches 79.4%mAP at 34.6 frame/s,and the mAP can raise to 82.9%after using the multi-scale testing strategy.

Multi-target range and velocity measurements of a digital phased array radar system

As the core of a digital phased array radar system,a radar signal processing environment is created to measure multitarget range and velocity information. The radar echo signal is achieved by superposing target echo, noise, clutter and jamming signals linearly. Considering that these signals have many types,two typical combinations are selected to construct the multi-target echo signal and the simulated echo signal is used as the input of the signal processing environment. This environment mainly adopts pulse compression,moving target indication and detection technologies to process the echo signal.It is found that the frequency domain method is more desirable for the pulse compression effect than the time domain method,and multi-target range information can be measured from the moving target indication result after using a double delay canceller. A new moving target detecting method is proposed,which can present the positive and negative velocity accurately with the multi-target range and velocity measured simultaneously. Simulation results indicate that the potential targets are detected from the chaotic radar echo signals successfully,and their range and velocity can be figured out correctly in the built radar signal processing environment.

An improved 3D shape haptic rendering algorithm for finger mounted vibrotactile device

To improve the sense of reality on perception, an improved algorithm of 3D shape haptic rendering is put forward based on a finger mounted vibrotactile device. The principle is that the interactive information and the shape information are conveyed to users when they touch virtual objects at mobile terminals by attaching the vibrotactile feedback on a fingertip. The extraction of shape characteristics, the interactive information and the mapping of shape in formation of vibration stimulation are key parts of the proposed algorithm to realize the real tactile rendering. The contact status of the interaction process, the height information and local gradient of the touch point are regarded as shape information and used to control the vibration intension, rhythm and distribution of the vibrators. With different contact status and shape information, the vibration pattern can be adjusted in time to imitate the outlines of virtual objects. Finally, the effectiveness of the algorithm is verified by shape perception experiments. The results show that the improved algorithm is effective for 3D shape haptic rendering.

Robust SLAM localization method based on improved variational Bayesian filtering

Aimed at the problem that the state estimation in the measurement update of the simultaneous localization and mapping(SLAM)method is incorrect or even not convergent because of the non-Gaussian measurement noise,outliers,or unknown and time-varying noise statistical characteristics,a robust SLAM method based on the improved variational Bayesian adaptive Kalman filtering(IVBAKF)is proposed.First,the measurement noise covariance is estimated using the variable Bayesian adaptive filtering algorithm.Then,the estimated covariance matrix is robustly processed through the weight function constructed in the form of a reweighted average.Finally,the system updates are iterated multiple times to further gradually correct the state estimation error.Furthermore,to observe features at different depths,a feature measurement model containing depth parameters is constructed.Experimental results show that when the measurement noise does not obey the Gaussian distribution and there are outliers in the measurement information,compared with the variational Bayesian adaptive SLAM method,the positioning accuracy of the proposed method is improved by 17.23%,20.46%,and 17.76%,which has better applicability and robustness to environmental disturbance.

A low-cost personal navigation unit

For the purpose of positioning in various scenes, including indoors, on open road, and side street buildings, a low-cost personal navigation unit is put forward. The unit consists of a low-cost MEMS(micro-electro-mechanical system) accelerometer, a gyroscope, a magnetometer and a GPS(global positioning system) chip, and it is capable of switching modes between indoor and outdoor situations seamlessly. The outdoor mode is MIMU(MEMS-inertial measurement unit)/GPS/magnetometer integrated mode and the indoor mode is MIMU/magnetometer integrated mode. The outdoor algorithm uses the extended Kalman filter to fuse data and provide optimum parameters. The indoor algorithm is dead reckoning, which uses vertical and forward accelerations to judge steps and uses a magnetometer to define heading. The two-axis acceleration data is used to calculate the adaptive threshold and estimate the confidence value of the steps, and when the confidence of both two axes data meet the conditions, the steps can be detected in the adaptive time windows. The detection precision is more than 95%. An experiment was conducted in complex situations. The experiment participant wearing the unit walked about 1 600 m in the experiment. The results show that the positioning error is less than 0.2% of the total route distance.

Efficient and reliable road modeling for digital maps based on cardinal spline

In order to realize an optimal balance between the efficiency and reliability requirements ofroad models,a road modeling method for digital maps based on cardinal spline is studied.First,the cardinal spline is chosen to establish an initial road model,which is specified by a series of control points and tension parameters.Then,in view of the initial road model,a gradual optimization algorithm,which can determine the reasonable control points and optimal tension parameters according to the degree of the change of road curvature,is proposed to determine the final road model.Finally,the proposed road modeling method is verified a d evaluated through experiments,and it is compared with the conventional method for digital maps based on the B-spline.The results show that the proposed method can resize a neaoptimal balance between the efficiency and reliability requirements.Compared with the conventional method based on the B-spline,this method occupies less data storage and achieves higher accuracy.

A calculation method for low dynamic vehicle velocity based on fusion of optical flow and feature point matching

Aming at the problem of the low accuracy of low dynamic vehicle velocity under the environment of uneven distribution of light intensity,an improved adaptive Kalman filter method for the velocity error estimate by the fusion of optical flow tracking and scale mvaiant feature transform(SIFT)is proposed.The algorithm introduces anonlinear fuzzy membership function and the filter residual for the noise covariance matrix in the adaptive adjustment process.In the process of calculating the velocity of the vehicle,the tracking and matching of the inter-frame displacement a d the vehicle velocity calculation a e carried out by using the optical fow tracing and the SIF'T methods,respectively.Meanwhile,the velocity difference between theoutputs of thesetwo methods is used as the observation of the improved adaptive Kalman filter.Finally,the velocity calculated by the optical fow method is corrected by using the velocity error estimate of the output of the modified adaptive Kalman filter.The results of semi-physical experiments show that the maximum velocityeror of the fusion algorithm is decreased by29%than that of the optical fow method,and the computation time is reduced by80%compared with the SIFT method.

Path prediction of flexible needles based on Fokker-Planck equation and disjunctive Kriging model

Path prediction of flexible needles based on the Fokker-Planck equation and disjunctive Kriging model is proposed to improve accuracy and consider the nonlinearity and anisotropy of soft tissues.The stochastic differential equation is developed into the Fokker-Planck equation with Gaussian noise,and the position and orientation probability density function of flexible needles are then optimized by the stochastic differential equation.The probability density function obtains the mean and covariance of flexible needle movement and helps plan puncture paths by combining with the probabilistic path algorithm.The weight coefficients of the ordinary Kriging are extended to nonlinear functions to optimize the planned puncture path,and the Hermite expansion is used to calculate nonlinear parameter values of the disjunctive Kriging optimization model.Finally,simulation experiments are performed.Detailed comparison results under different path planning maps show that the kinematics model can plan optimal puncture paths under clinical requirements with an error far less than 2 mm.It can effectively optimize the path prediction model and help improve the target rate of soft tissue puncture with flexible needles through data analysis and processing of the mean value and covariance parameters derived by the probability density and disjunctive Kriging algorithms.

Metrics analysis of tactile perceptual space based on improved NMDS for leather textures

To solve the fuzzy and unstable tactile similarity relationship between some sample points in the perception experiment,an improved non-metric multidimensional scaling(INMDS)is proposed in this paper.In view of the inconsistency of each sample s contribution,the maximum marginal decision when constructing the perception space to describe the tactile perception characteristics is also proposed.The corresponding constraints are set according to the degree of similarity,and controlling the relaxation variable factor is proposed to optimize the perception dimension and coordinate measurement.The effectiveness of the INMDS algorithm is verified by two perception experiments.The results show that compared with the metric multidimensional scaling(MDS)and non-metric multidimensional scaling(NMDS)algorithms,the perceptual space constructed by INMDS can more accurately reflect the difference relationship between different leather sample points perceived by people.Moreover,the relative position of sample points in the perceptual space is more consistent with subjective perception results.

Detection of solar radio burst intensity based on a mod ified multifactor SVM algorithm

To realize the automatic detection of solar radio burst(SRB)intensity,detection based on a modified multifactor support vector machine(SVM)algorithm is proposed.First,the influence of SRB on global navigation satellite system(GNSS)signals is analyzed.Feature vectors,which can reflect the SRB intensity of stations,are also extracted.SRB intensity is classified according to the solar radio flux,and different class labels correspond to different SRB intensity types.The training samples are composed of feature vectors and their corresponding class labels.Second,training samples are input into SVM classifiers to one-against-one training to obtain the optimal classification models.Finally,the optimal classification model is synthesized into a modified multifactor SVM classifier,which is used to automatically detect the SRB intensity of new data.Experimental results indicate that for historical SRB events,the average accuracy of SRB intensity detection is greater than 90%when the solar incident angle is higher than 20°.Compared with other methods,the proposed method considers many factors with higher accuracy and does not rely on radio telescopes,thereby saving cost.

Modeling the special intersections for enhanced digital map

A new lane-level road modeling method based on cardinal spline is proposed for the special intersections which are covered by vegetation or artificial landscape in their central regions.First,cardinal spline curves are used to fit the virtual lanes inside special intersections,and an initial road model is established using a series of control points and tension parameters.Then,the progressive optimization algorithm is proposed to determine the final road model based on the initial model.The algorithm determines reasonable control points and optimal tension parameters according to the degree of road curvature changes,so as to achieve a balance between the efficiency and reliability of the road model.Finally,the proposed intersection model is verified and evaluated through experiments.The results show that this method can effectively describe the lane-level topological relationship and geometric details of this kind of special intersection where the central area is covered by vegetation or artificial landscape,and can achieve a good balance between the efficiency and reliability of the road model.