Jestr
JestrJOURNAL OFJournal of Engineering Science and Technology Review 10 (1) (2017) 168- 172Engineering Science andTechnology ReviewResearch Articlewww.jestr.orgFacial Animation of Life-Like Avatar based on Feature Point ClusterSamuel Gandang Gunanto1,2 *, Mochamad Hariadi1, Eko Mulyanto Yuniarno1 and Matahari Bhakti Nendya21Department of Electrical Engineering, Institut Teknologi Sepuluh Nopember Surabaya – Indonesia2Department of Animation, Institut Seni Indonesia Yogyakarta – IndonesiaReceived 3 March 2016; Accepted 15 January 2017AbstractThe problems that are often encountered in the making of facial expressions on 3D animated avatars heavily dependenton the ability of key animators to create movement and combine them into a series of facial motion animation. Thisresearch proposes an automatic preprocessing for facial animation using feature point cluster. The mechanism of atransfer system for animated facial expressions using a radial basis function technique that is effective, especially interms of production speed. This study focuses on the improvement of animated facial expressions using the feature pointcluster. The result is to provide improved motion animation more expressive. This process will be easier and faster forthe formation of facial motion animation without the need for process adjustments manually.Keywords: facial animation, avatar, feature point cluster, retargeting, radial basis function.1. IntroductionOne form of the use of digital characters is avatars. Avataracts as a personal representative of a person in the digitalworld. Personal representation form by using avatars can bea person, robots, superheroes, cartoon characters or animals.Avatar on the genre game Real- Time Strategy (RTS) isrepresented in the form of the face of the characters used.The forms of the character face are very closely linked tostrengthening the presence of facial animation for characters.Traditional approaching to the formation of animatedfacial expressions models rely on an artist to create amovement key and then combine them into a series ofmovements of facial expression. A game with the rules ofthe interactive game requires the presence of facialanimation in their communication or interaction with thegame area. Due to limited resources, in-game facialanimations are often denied. But lately with the advent ofgameplay system of animated films interspersed thereinrequire the presence of animated facial expressions are thereto provide entertainment aspect and the story for the player.Problematic encountered for manufacturers is the use oftemplates and the same facial movements in differentmodels is a very time-consuming task for the animators [1].These problems also occur in the manufacture of facialanimation motion because the adjustment of naturalmovement and detail needed to make a facial expression beattractive and make the characters come alive. Hence thesearch for a solution that is capable of handling the issue atop priority to be tackled, especially which is able to bring adesign appropriate technology in the animation industryE-mail address: gandang6@gmail.comISSN: 1791-2377 2017 Eastern Macedonia and Thrace Institute of Technology. All rights reserved.sector.This study proposes a method of facial animation usingfeature point cluster approach. Mesh deformation byconsidering the feature-point cluster is expected to providethe model results avatars are able to visualize the expressionwithout the need for a conversation or dialogue.2. State of the Arts2.1 Computer Facial AnimationLanguage expression was as developed by Hjortsjo in 1970was one of the first attempts to systematize the mapping offacial expressions by facial muscle activity was observedfrom various differences in facial expressions of the actors[2]. Facial Action Coding System, or commonly known asFACS as the development of methods of expressionlanguage in 1978. FACS divides the face area to 66 ActionUnit (AU) were able to reconstruct the face with variousexpressions naturally [3]. FACS system is still often used byanimators to reconstruct facial expression by utilizing thecontrollers in the expression of facial animation systemdescriptively.Mapping technique in the movement of facial expressionin animation system can be guided based on the marker. Themarker-based system is a standard that has been developedcommercially or fabricated. This system is used because ofthe lack of computing used taking into account the speed ofthe process is instantaneous or real-time.2.2 Deformation-Based ApproachesDeformation directly defined on the surface of the facialmesh that often produce the quality of animation. Thisignores the facial anatomy and structure of the existingmuscle. So that the focus is only on making variations of
Samuel Gandang Gunanto, Mochamad Hariadi, Eko Mulyanto Yuniarno and Matahari Bhakti Nendya/Journal of Engineering Science and Technology Review 10 (1) (2017) 168- 172facial expression by manipulating thin-shell mesh. Thiscategory includes morphing between different models andsimulated artificial muscles in shape splines [4], mesh [5], orfree-form deformations [6].3. Experimental Design2.2.1 Morphing 2D and 3D MorphingBeier and Neely [7] shows the 2D morphing between twoimages that are manually selected features associated with aline. Warp function conducted based on where the affectedand surrounding areas. Realistic approach of this methodrequires manual interaction in the form of coloring, selectionof features correspondent, setting the parameters of warp anddissolve. Variations can be applied in the viewpoint of thedestination image or the selection of correspondence feature.Head movement is very difficult in the synthesis process ifthe feature of interest is closed. To overcome this, Pighin etal. [8] combine 2D with 3D transformations morphinggeometry model. They animate the facial expressionanimation keys with 3D geometry interpolation when theimage started morphing appropriate texture. This approachlooked realism, although the animation is limited tointerpolate between key facial expression has beendetermined in advance.2D and 3D morphing method can produce good facialexpressions and quality, but have limitations similar tointerpolation approach. Selection point relating to the imageof the destination is still done manually and depending onthe angle of view and does not have the general form for adifferent face. In additional viewpoint animation also, betied to a specific destination image only.Fig 2. Facial Animation of Life-Like Avatar based on Feature PointCluster3.1 Facial RiggingFacial rigging is the process of making a control point forfacial animation, this usually made by the animator. In thisstudy, the facial rigging based on the location of facialfeature points on the face mask that refers to the approachused in the FACS optitrack motion capture system.Facial rigging process is done manually on each modelof the avatar. The result of this process is a data point of theface mask features from avatar models, see Fig. 3. Propfacial feature point coordinates will be used as a centroid inthe process of clustering and is used in the process ofretargeting.2.2.2 Free Form DeformFree Form Deformation (FFD) volumetric change the objectby changing the control point in three-dimensional cubiclattice [9]. In concept, a flexible object embedded in a cubiclattice-shaped 3D control grid and is composed of dots ofcontrol. Box cubic lattice of control capable of adjusting thechanges done to the object so that the object is modifiedtherein remains in the square lattice, see Fig. 1.Fig. 3. Facial Rigging on Target Face Models3.2 ClusteringClustering process used in this study refers to a clusteringalgorithm k-nearest neighbor with some modification in theprocess of definition value of k. Algorithm k-nearestneighbor (k-NN) is a method to perform the clustering ofobjects based on the learning process from data that werelocated closest to the object. In this case, learning data is thedata of vertices which are located close to the point features.The process of modified clustering using k-NN to find afeature point cluster can be seen in the following flowchart,see Fig 4.Fig 1. Free Form Deformation (FFD) [2]2.3 Facial Motion Capture DataFacial motion capture data used in some motion captureanimated film. In motion capture obtained and cleaned bythe movement of the animation. An array of good qualitycamera catches the data used to build a 3D marker locationin the area of the face. Although the optical system is verydifficult to use and expensive, provide reconstructed datatimeliness and motion information. Once the data is ready,facial animation can be created by processing structure [10]muscle or blend shapes [11].169
Samuel Gandang Gunanto, Mochamad Hariadi, Eko Mulyanto Yuniarno and Matahari Bhakti Nendya/Journal of Engineering Science and Technology Review 10 (1) (2017) 168- 172with weights 𝑊! 𝑤!! 𝑤!! 𝑤!! ! . So as to solve thesystem we calculate the value 𝑊! 𝐻 !! 𝑇! . RBF networkonce trained for each axis, position in space 𝑡 targets foreach point 𝑠 of source space is obtained by applying thetransform 𝐹 𝑠 is generated.3.4 Mesh Deformation (Skinning)Linear blend skinning (LBS) purposed for changing the meshaccording to the local transformation of the skeleton.Deformation skinning techniques adapt to the changinglocation of point features, the feature point position is alsoconsidered as a point of joint motion. Weights for each pointdefined 𝑤! 1 weight to bone- i 𝑤! [0,1]. Weightingwith a value of 0 means that the feature does not affect thepoint vertex in the mesh point, whereas if the weight value of1 means that the point vertex in the mesh is affected only bythe feature point only. Vertex position 𝑣! the frame f isdefined by Eq. 5.𝑣! 𝑣! Fig. 4. Flowchart Proposed Method of Modified k-NN Based on FeaturePoint Cluster!!!! 𝑤!with ℎ𝑠! 𝑠! ℎ 𝑠! 𝑠!𝑠! 𝑠!4.1 Feature Point ClusterFeature point cluster formation performed by the clusteringmodel involving all the target face vertex points that existwithin the target model's face mask. Determined 33 pointsvertex associated with facial feature points as the center ofthe cluster or centroid. Calculations the similarity done byusing the Euclidean distance to get the membership of eachcluster. The cluster results obtained then used in the processof determining the affected area of movement.(1)(2)! 𝑠𝑐!!(multi-quadricfunction) and 𝑠𝑐! 𝑚𝑖𝑛!!! 𝑠! 𝑠! .The training network consists of the completion of 3linear system of N (in the case of 3 dimensional) like Eq. 3.𝑡! 𝐹 𝑠!Fig. 5. Feature Point Cluster Processing.In this study, the process of using the clusteringalgorithm k-nearest neighbor (k-NN), which is a method toclassify the object based on the learning data that waslocated closest to the object, are modified using featurepoint, see Fig.4. The value of k in the k-NN is determined bythe position of the vertex associated with facial featurepoints that have been determined on the facial rigging.(3)Suppose H is a matrix like 𝐻!" ℎ 𝑠! 𝑠!and𝑇! 𝑡!! 𝑡!! 𝑡!! ! . So by using Eq. 3, the system can bedefined as:𝑇! 𝐻 𝑊!(5)4. Result and Discussion3.3 Radial Basis FunctionRadial Basis Functions (RBF) is often used in computergraphics applications for estimation process and surfaceinterpolation. In this study, RBF is used as a transformationof space. To do this first defined two rooms with two featurepoint set. Suppose S0 as the set point of the source feature,T0 as the target feature point set, and N is the size of the set.Each controls point 𝑠! 𝑆! had a relationship with 𝑡! 𝑇! .After training with two pieces of the feature point set, RBFcan transform the position of the source space to the targetspace with RBF formulation in Eq. 2.𝐹 𝑠! 𝑑!"Where N is the number of feature points and 𝑑!" is achange in the location of the i-th feature point on the frame f.A major component in the clustering algorithm ismeasuring the distance between the vertex and point featuresand the decision to make a cluster. Data in the form of thesame vectors are able to use a euclidean distance matrixapproach to measuring the distance between vertex and pointfeatures, see Eq. 1. The decision for clustering process basedon the similarity factor between vertex to their centroid.d ( x2 x1 )2 ( y2 y1 )2 ( z2 z1 )2!!!! 𝑤!(4)170
Samuel Gandang Gunanto, Mochamad Hariadi, Eko Mulyanto Yuniarno and Matahari Bhakti Nendya/Journal of Engineering Science and Technology Review 10 (1) (2017) 168- 1724.2 Radial Basis Function Space TransformTransformation of space using radial basis function is usedas a way to transform the space between the facialexpressions, based on facial motion capture data, and thetarget face model, the avatar. Morphological forms of humanfaces, in general, are very varied and have differences withthe morphological form of 3D characters such as cartooncharacters, monsters or animals. This resulted in themovement of facial feature points that use humans as asource of animation data can’t be directly used. There areseveral things to consider in the use of animation data suchas scale and orientation.At frame 0, the data displacement is 0, this is because theframe 0 is the origin frame for feature point. In the nextframe (1-7) the displacement of the frame, which indicatesthe movement of feature points describe in Fig. 8.FP 1FP 2FP 6FP 10FP 3FP 4FP 9FP 11FP 22FP 24FP 26FP 290.10.05Fig. 6. Radial Basis Function Result to Avatar Face Model0To determine the feature points displacement on the faceof the avatar, the feature point displacement calculation fromthe RBF conducted. As the discussion and observation,taken 12 points from the character models considered torepresent the avatar character formation of facialexpressions. Fig. 7 presents 12 feature points are used as adiscussion.1234567Fig. 8. Feature Points Displacement Graph of The Avatar's Face Model.From the graph in Fig 8 is known that the displacementof facial feature points on the avatar’s face is well formed.The process of moving facial feature points occurs linearlyand depend on the number of mesh faces and how themorphological form of the model used. The morphologicalform is similar to a human and visually good recognizedclearly, see Fig 9.The level of conformity result of transformationprocesses in the model space RBF human face character canbe calculated using the standardized approach like standarddeviation of the facial motion capture data with thetransformed data RBF. The result of the calculation ofstandard deviation for all the feature points can be seen inTab. 2. The results of transformation of space in the face ofthe character models have a standard deviation of 0.0034.Fig. 7. Feature Point Used For Analyzed On Avatar FaceModelTable 2 The Standard Deviation of Feature PointsDisplacement.Fig. 7 is a position of the feature point of the avatar'sface. The process of space transformation RBF providesresults for each point feature that has undergone a process oftransformation.Data observed on seven frames from the data of facialmotion capture data and the results of the transformationprocess RBF space. Facial motion capture data is used as areference to determine the transfer point features. When thefeature points of facial motion capture data are changed,then all the feature points on the avatar’s face moving inaccordance with the data source. The Changes of featurepoints from the result of the transformation of space in theavatar’s face are presented in the Tab. 1.Feature Point123469101122242629Average Standard DeviationTable 1 Results of The Displacement Feature Points on TheAvatar Face171Standard 10.02280.02470.02990.02550.03140.0034
Samuel Gandang Gunanto, Mochamad Hariadi, Eko Mulyanto Yuniarno and Matahari Bhakti Nendya/Journal of Engineering Science and Technology Review 10 (1) (2017) 168- 1724.3 Mesh DeformationMesh deformation process begins by determining the weightof each vertex are interconnected in a cluster member. Theweighting of point refers to the displacement of the vertexwhen the process of movement occurs. The process ofdetermining the point weighting effect on the formation ofan animation on a mesh face is calculating using LBS. Tab.3 are an example of the weight calculation results for eachvertex.Table 3. Example of The Weight Calculation Results forEach VertexNoVertex CoordinatesWeightxyz1-0.05590.216127 0.0574650.6387482-0.055640.224933 0.0644670.7401253-0.055640.224933 0.0644670.7401254-0.055640.224933 0.0644670.7401255-0.05590.216127 0.0574650.6387486-0.05590.216127 0.0574650.6387487-0.055640.224933 0.0644670.7401258-0.067090.226293 0.0784950.7960499-0.067090.226293 0.0784950.79604910-0.060020.215447 0.0753910.684178Fig. 9. Results of Facial Animation on The Avatar’s Face Models5. Conclusion And Future WorksReuse of animation techniques (retargeting) using thetransformation of the space-based Radial Basis Function(RBF) produces character expressions that follow themovements of the actors and the expression of the sameapproach with the actors. RBF space transformation is atechnique with a linear approach that does not require acomplicated computation. Mechanical merging with meshmodels features point on the stages of deformation meshprovides a visual form of the transformation result.Transformation of space RBF provides solutions in theformation of facial animation in accordance with themovement of the actors and relies on the need to control thebone system, blend shape weight, and other parameters inthe formation of specific facial animation.The complexity of this research can be improved bybuilding a transfer system for animated facial expressions inreal time. The transformation process of facial expressionscan also be tested with other approaches, such as usinggeodesic distance where considering the measurements ofthe facial surface.Vertex weight calculation results in Tab. 3 was used as areference in determining the new vertex position inaccordance with the process of the formation of anexpression. Fig. 9 is presented the process of meshdeformation on the avatar’s face model, especially at thepoint features that affect the formation of animated facialexpressions.This is an Open Access article distributed under the terms of theCreative Commons Attribution LicenceReferences1.2.3.4.5.6.Orvalho, V. C., Zacur, E., & Susin, A. Transferring FacialExpressions to Different Face Models. Ibero-AmericanSymposium on Computer Graphics - SIACG. The EurographicsAssociation. (2006).Hjortsjo, C. Man's Face and Mimic Language. Malm: NordensBoktryckeri. (1970).Ekman, P., & Friesen, W. V. Facial Action Coding System: atechnique for the measurement of facial movement. Palo Alto:Consulting Psychologists Press. (1978).Viad, M., & Yahia, H. Facial animation with wrinkles. Proceedingsof the Third Eurographics Workshop on Animation andSimulation. (1992).Singh, K., & Fiume, E. Wires: A geometric deformation technique.SIGGRAPH Proceedings. pp. 405–414. (1998).Kalra, P., Mangili, A., Thalmann, N., & Thalmann, D. Simulationof facial muscle actions based on rational free from deformations.Eurographics, vol 11. pp. 59–69. (1992).7.Beier, T., & Neely, S. Feature-based image metamorphosis.SIGGRAPH Proceedings. ACM Press. pp 35-42. (1992).8. Pighin, F., Auslander, J., Lischinski, D., Salesin, D., & Szeliski, R.Realistic facial animation using image-based 3d morphing. (1997).9. Sederberg, T., & Parry, S. Free-form deformation of solid geometrymodels. Computer Graphics, SIGGRAPH, volume 20. pp 151–160. (1996).10. Choe, B., Lee, H., & Ko, H. (2001). Performance driven musclebased facial animation. The Journal of Visualization andComputer Animation, 12(2). pp 67-79. (2001).11. Deng, Z., Chiang, P., Fox, P., & Neumann, U. Animating blendshape faces by cross-mapping motion capture data. Proceedings ofthe 2006 symposium on Interactive 3D graphics and games. pp 4348. (2006)
category includes morphing between different models and simulated artificial muscles in shape splines [4], mesh [5], or free-form deformations [6]. 2.2.1 Morphing 2D and 3D Morphing Beier and Neely [7] shows the 2D morphing between two images that are manually selected features associated with a line.
casing milling and reamer milling, which simplifies the operation process, improves operation efficiency and . 3 is 0.9mm, α is 30 , 8 . is connected to the diverter to form a high-pressure chamber
2 DRDO, Delhi, India Received 6 April 2018; Accepted 17 February 2019 _ Abstract The brief review of methods used for estimating the target state in single and multi-sensor bearing only tracking (BOT) is presented in this paper. It deals with the target state estimation using bea
2. The Theory of creep in Steel Steel for power plant are often placed in service at elevated temperatures and exposed to static mechanical stresses (e.g., turbine rotors in steam generators that experience centrifugal stresses, and high-pressure steam lines) [12]. The steam turbine casing attains the running steam temperature during
piezoelectric transducer, a suitable inductance would be enough in EIMN. As for a broadband piezoelectric transducer (BPT), however, methods of designing EIMN are usually complicated and hard to understand. Some successful methods [3][4] of designing EIMN have been developed, but it is still meaningful to find a new way
14 Mercedes Benz A250 2014 Wheel alignment Wheel alignment specifications. Contacted VACC, no information available, checked online for OEM websites, blocked from USA and Europe. 15 Suzuki Jimny 2008 Collision repair Body repair quarter panel replacement procedures. Contacted VACC, no information available, checked online for OEM websites, blocked from USA and Europe. 16 LDV G10 2016 Service .
Part 2, Design – High Strength Steels was combined with Part 1, Design in 1993. Part 5, Special Types of Construction was combined with Part 1, Design in 2008. Part 10, Bearing Design, and Part 11, Bearing Construction, were combined into a new Part 5, Bearing Design and Construction in 2013.
Belajar sejarah melalui e-learning bukan menggantikan proses pembelajaran konvensional, namun pembelajaran melalui e-learning ini untuk memperkuat nilai belajar konvensional dengan menggunakan teknologi internet. Maka dalam Kamarga (2002, hlm 55) menyebutkan : Belajar sejarah melalui e-learning dapat diartikan sebagai upaya
Synthetic biology advances science & technology while mitigating danger –General capability to engineer biological systems –Increased speed and scope of response to threats Systems ApplicationsApplication Parts & Fabrication Design Biology is a technology for processing information, materials, and energy.
