Research On Key Technology Of Virtual Scene Rendering Based On Visual Perceptual Optimization

With the development of computer technology,the rise of big data and internet research applications,the demand for immersive sense of virtual scene is constantly improved.The virtual reality system emerges a lot of special points,such as the more number of 3D entities,the high complexity and the large scale of the scene,in scene generation and exploration.The traditional modeling and rendering methods cannot solve the outstanding contradiction between virtual reality system in terms of model complexity,rendering realism and scene scale.Therefore,under the constraints of computer hardware resources,it has important theoretical significance and application prospect to study visual perception optimization based virtual scene modeling,rendering and exploration technologies.This paper analysis and research on several key technologies visual perceptual degradation of simplified 3D model based on multi-viewpoint images and multi-resolution modeling of 3D model,virtual scene viewpoint quality evaluation and intelligent optimal viewpoint selection,intelligent combinatorial optimization of model resolution level for 3D virtual scene,automatic exploration of large-scale virtual scene and the fidelity evaluation of virtual ocean environment.The main research results and innovation are as follows:1)Aiming at the problem that the viewpoint number in the image based quality evaluation of mesh simplification is selected by experience and the evaluation factors were inadequate.Thus,a multi-viewpoint image based visual degradation evaluation of simplified mesh model was proposed.First,the candidate viewpoint number was determined through the analysis of the relation between the viewpoint number and the efficiency in the field of mesh simplification and the optimal viewpoint selection of the scene.The image quality metrics were selected either known for efficiency or for widespread use in the community from three categories statistics based,information based and human visual system based classes.Then,the multi-viewpoint image evaluation framework was established.The evaluation performance of each evaluation factors were obtained by the correlation statistical analysis between the objective evaluation results and the subjective mean opinion score in the The IEETA Simplification Database,and provided the evaluation factors selection strategy for mesh simplification visual degradation evaluation.On this basis,the visual difference between the original model and the simplified model under the optimal viewpoint set was considered as the visual perception error caused by the simplification.Through combining QEM algorithm with triangle feature factor constraints,a visual and geometric multi-featuremetric function was constructed to control the simplified depth of the mesh model,then ensure that the visual and geometric features of the simplified model can be maintained.The comparison experiment results verify the effectiveness of the method.2)The traditional viewpoint quality evaluation metrics either requires long computation time or fails to provide results which consistent with human observation habits.Aiming at this problem,two metrics were proposed respectively from image and scene viewpoint attribute aspects.First,through analyzing the influence of several digital image factors(luminance,chrominance,texture details and spatial location.etc.)on human visual characteristic,the mathematical models were established.Calculating the visual perception influencing factors,the weighted value of each equal-sized scene image region was generated after normalization.Combined with information entropy theory,a viewpoint quality evaluation metrics named regional perceptual weighted information entropy(RPWIE)was proposed.Secondly,the scene view descriptor was defined from the geometric features,semantic features and image features of the scene.The vertex curvature which represent the visual feature of the model was used to construct the surface visibility geometry feature measurement function model named surface saliency entropy.By integrating the viewpoint geometry information with the viewpoint image information,a viewpoint quality evaluation metric viewpoint potential(VP),based on viewpoint attribute fusion was proposed.Finally,AW-PSO algorithm was utilized to select the optimal viewpoint automatically and intelligently.The validity of the two proposed algorithms was verified by the example of virtual sea war environment and compared with the existing algorithms.The results show that the RPWIE algorithm is efficient,and the best viewpoint image is calculated to fit the human visual perception.The VP algorithm calculates that the optimal viewpoint can contain more scene information,satisfy the observer's need for the scene,and the observation angle is more comfortable.3)The traditional design process of large-scale 3D scene is cumbersome,lengthy and fails to provide ideal scene rendering effect.To this,a feedback mechanism based level of detail(LOD)model combinatorial optimization for 3D scene was proposed.Through analyzing the influence of observation distance,moving speed and other factors on visual perception,the mathematical model of visual sensitivity was established.Through frame rate and visualization effect feedback,invoking the model level selection criterion to adjust the model's LOD level,and eventually an optimal scene model combination solution could be obtained.Experimental results demonstrate that the proposed method can effectively reduce the number of combinatorial evaluation,and the higher fidelity 3D scene generated by theoptimal model combination solution can be quickly obtained.Simultaneously,an intelligent LOD model combinatorial optimization algorithm was proposed to improve the scene visualization effect and the optimization speed in LOD model Particle Swarm Optimization(PSO)optimization based on image information entropy.First,scene mutil-feature information fusion evaluation intelligent combination of LOD model framework was built.Second,modified PSO method which was strengthened by Genetic Algorithm(GA)was utilized to improve the searching precision of the optimal solution and optimization speed.The results show that the method is better than the single index evaluation method under the same experimental conditions both in the optimization efficiency and scene visualization effect.4)Aiming at the shortcomings of large-scale scene exploration methods,such as unintuitive scene information description,long time-consuming exploration,uneven viewing angle and unsmooth path.An automatic large-scale virtual scene exploration method was proposed.Firstly,the scene was adaptively divided into several meaningful and easily analyzed sub-regions according to the optimal view distance criterion.The optimal viewpoint sphere of each region was constructed by convex hull.Secondly,viewpoints were sampled on the sub-region viewpoint sphere and evaluated,the optimal viewpoint set for each sub-region was computed through clustering and greedy algorithm.Then,GA-TSP and ACO-TSP algorithm were used to optimize the viewpoint ordering of intra-region and the exploration order of inter-region respectively.Finally,the three Hermite curve was utilized to smooth the inflection point in the exploration path.The experimental results show that the proposed algorithm can effectively generate an automatic smooth and non-intersecting path,and with the characteristics of good roaming comfort,strong immersion and high scene information recognition.5)In order to achieve effective evaluation the visual fidelity of the virtual ocean environment corresponding to different ocean wave modeling algorithms,a subjective and objective combination based virtual marine environment visualization fidelity assessment was proposed.On the basis of following the basic principles of 3D visual simulation real-time,consistency,interactivity and combination,combining with information perception and cognitive principles of human visual system(HVS).From virtual ocean environment description,viewpoint observation,scene cognition and dynamic display aspects,the composition and relationship of the factors affecting the visual simulation fidelity were analyzed,and the diversity of evaluation indicators was realized.The analytic hierarchyprocess(AHP)and entropy weight methods were used to determine the subjective and objective weights respectively.The game theory was used to calculate the mixed weights to take into account the subjective attribute preferences and reduce the subjective randomness,so the index weight can reach the subjective and objective unity.Finally,the improved radar chart method was applied to realize the visualization and quantification of the evaluation result.The proposed model was applied in three typical ocean environment visualization.The experimental results show that the method can reasonably and correctly solve the problem of visual fidelity evaluation,and provide a new perspective for the fidelity evaluation of large-scale complex visual simulation system.