3D voxel model reconstructed from images often suffers from problems of noise and incomplete building structure.Layered contour fitting method provides a way to solve this kind of problem.However,the cross planar contours are very complicated in the realistic scene.In view of this,this paper proposes a two-step contour fitting method to fit the cross-sectional contours of a building on the layer image.The hierarchical projection method is used to project the initial 3D model into a two-dimensional layer image which is then de-noised by a contextual density-based clustering algorithm.Precise fitting results of planar contours are obtained by shape classification and shape fitting.The final 3D model is composed of the results from the upper edge of the layer image.Experimental result shows that compared with the original reconstruction model,the precise building model has more regular and complete shape with less noise,and substantially reduces storage space.

The 3D virtual tour system often provides immersive tour experience.It cannot offer information service about tour and is hard to fulfill users’ personalized requirement.To tackle this issue,an indoor positioning method based on IBeacon Bluetooth communication technology with low consumption is proposed to provide indoor positioning and automatic tour-guidance service in indoor culture tourist facilities.It is based on RSSI ranging principle and plane projection-based 3D space positioning technology,and supports receiving transmission signals from IBeacon Bluetooth base station by mobile phone to realize fast positioning.This paper applies IBeacon-based indoor positioning method in online 3D virtual tour system,and establishes an online-to-offline cultural tour guidance system combining virtuality with reality,which provides tourists with online virtual experience,offline realtime positioning and location-based information service.Application results show that the proposed system has higher speed of indoor positioning and accuracy fulfilling application requirement.It can provide users with better travel experience.

This paper proposes an improved virtual view point rendering algorithm to solve the problems of ghosting and holes in the virtual view point rendering.This method marks the unreliable regions of depth map based on the depth information.Through multi-threshold image segmentation,the image is divided into different parts.A virtual view point is generated by making use of 3D image warping.In order to deal with the small cracks,local median filtering is used to obtain a single-view point image.Two generated virtual images are blended together to eliminate the disocclused holes in the desired view.An inpainting method is utilized to eliminate the remaining holes.Experimental results show that high quality virtual view points are generated,the Peak Signal to Noise Ratio(PSNR) of virtual view point can be 0.74 dB higher than that standard reference software of virtual view point rendering.

To solve the shortage or low utilization of lack of experiment teaching equipment in some universities,this paper proposes a virtual/remote experiment system of double tank.It is based on Networked Control System Laboratory(NCSLab) architecture and accomplished by combining 3D virtual reality and remote control technology.3D modeling and rendering,mathematic modeling and system integration are included in the process of designing the experiment system.Coupled-tank experiment result shows that the system is able to share experiment equipment and bring students a strong sense of immersion.Also,the results of virtual and remote experiments can be mutually verified and the personalized monitoring interface also brings a better user experience.

Fast and correct detection of on-road objects is a difficult problem in the field of assistant driving and automatic driving of intelligent vehicles.So,based on overall visual structure model,the visual cognitive mechanism from the local to the whole is simulated.The structural model is introduced in the object part model,as well as the statistical model,the overall visual structural model with scale invariance and posture invariance is gained.Through weak-labeled sample image,the model can automatically label the representative local areas so that this model can be more discriminable in detection.The new model is adapted to detect obstacle object in road area.Experimental results show that the object detection algorithm based on the proposed model has higher efficiency than traditional algorithms.It has high detection rate and low error rate.

Aiming at the low precision of current location based applications in tourism area,considering that map matching algorithms are mainly intended for intelligent transportation systems,a score rule-based dynamic map matching algorithm is proposed.The algorithm takes road width into account and extends the line road network modeling.Based on probabilistic map matching methods,the dynamic threshold mechanism is adopted to fit the complex location environment.Closeness,topological relation as well as dynamic confidence area are applied to figure out the candidate roads.It uses cumulative score method to choice road.Experimental result indicates that,the algorithm is used in the campus navigation application,and the accuracy of roads identification reaches 89.3%,which is 18.4% higher than that of the nearest point-curve distance method and 10% higher than current map applications.The proposed algorithm can effectively improve the accuracy of road selection and location,meanwhile improves map display.

Aiming at the disadvantages that the existing modeling methods of human motion cannot be applied at the general nonlinear and non-Gaussian cases,based on the Markoff model,an improved modeling method of human motion is proposed.The Markov process of human motion prediction is presented by using the Auto Regressive(AR) tree,and the paper proposes extensions to AR trees and introduces the Dynamic Forest Model(DFM) and describes its training,regularization and the realization of the accurate modeling of human motion.Example research results show that performance of DFM is better than other benchmark algorithms,Hidden Markov Model(HMM) and Gaussian Process Dynamical Model(GPDM),and computing efficiency of the proposed methods is high.

Recently,researchers in the field of information retrieval have proposed many ranking algorithms supporting search result diversification,but there is no literature on the performance analysis and comparison of these algorithms.Based on a traditional data fusion ranking algorithm CombSum,this paper proposes a new ranking algorithm CombSumDiv,which considers relevance and diversity of the documents at the same time.Besides,a comparison among CombSumDiv and two explicit ranking algorithms xQuAD and PM2 is conducted.Experiments are carried out with query datasets and ClueWeb09B datasets from the TREC diversity task.The result shows that the query performance of CombSumDiv algorithm is better than that of xQuAD and PM2 algorithm,and these three algorithms have strong stability and anti-interference ability.