Cesium源码剖析---Ambient Occlusion(环境光遮蔽)
Ambient Occlusion简称AO,中文没有太确定的叫法,一般译作环境光遮蔽。百度百科上对AO的解释是这样的:AO是来描绘物体和物体相交或靠近的时候遮挡周围漫反射光线的效果,可以解决或改善漏光、飘和阴影不实等问题,解决或改善场景中缝隙、褶皱与墙角、角线以及细小物体等的表现不清晰问题,综合改善细节尤其是暗部阴影,增强空间的层次感、真实感,同时加强和改善画面明暗对比,增强画面的艺术性。AO简单来说就是根据周围物体对光线的遮挡程度,改变明暗效果。AO具体理论原理在网上都可以找到,感兴趣的可以去查阅,此处只把Cesium中AO的实现过程作一下介绍。闲言少叙,直接上代码。
1. 开启AO及效果
在Cesium中开启AO效果非常简单,和之前讲的开启Silhouette效果类似,代码如下:
1 var ambientOcclusion = viewer.scene.postProcessStages.ambientOcclusion; 2 ambientOcclusion.enabled = true; 3 ambientOcclusion.uniforms.ambientOcclusionOnly = false; 4 ambientOcclusion.uniforms.intensity = 3; 5 ambientOcclusion.uniforms.bias = 0.1; 6 ambientOcclusion.uniforms.lengthCap = 0.03; 7 ambientOcclusion.uniforms.stepSize = 1; 8 ambientOcclusion.uniforms.blurStepSize = 0.86;
没有开启AO效果如下图一,开启AO效果如下图二,单纯的AO图如图三:
2. JS内部代码实现
在PostProcessStageLibrary类中添加AO功能的代码如下:
1 PostProcessStageLibrary.createAmbientOcclusionStage = function() { 2 var generate = new PostProcessStage({ 3 name : 'czm_ambient_occlusion_generate', 4 fragmentShader : AmbientOcclusionGenerate, 5 uniforms : { 6 intensity : 3.0, 7 bias : 0.1, 8 lengthCap : 0.26, 9 stepSize : 1.95, 10 frustumLength : 1000.0, 11 randomTexture : undefined 12 } 13 }); 14 var blur = createBlur('czm_ambient_occlusion_blur'); 15 blur.uniforms.stepSize = 0.86; 16 var generateAndBlur = new PostProcessStageComposite({ 17 name : 'czm_ambient_occlusion_generate_blur', 18 stages : [generate, blur] 19 }); 20 21 var ambientOcclusionModulate = new PostProcessStage({ 22 name : 'czm_ambient_occlusion_composite', 23 fragmentShader : AmbientOcclusionModulate, 24 uniforms : { 25 ambientOcclusionOnly : false, 26 ambientOcclusionTexture : generateAndBlur.name 27 } 28 }); 29 30 var uniforms = {}; 31 defineProperties(uniforms, { 32 intensity : { 33 get : function() { 34 return generate.uniforms.intensity; 35 }, 36 set : function(value) { 37 generate.uniforms.intensity = value; 38 } 39 }, 40 bias : { 41 get : function() { 42 return generate.uniforms.bias; 43 }, 44 set : function(value) { 45 generate.uniforms.bias = value; 46 } 47 }, 48 lengthCap : { 49 get : function() { 50 return generate.uniforms.lengthCap; 51 }, 52 set : function(value) { 53 generate.uniforms.lengthCap = value; 54 } 55 }, 56 stepSize : { 57 get : function() { 58 return generate.uniforms.stepSize; 59 }, 60 set : function(value) { 61 generate.uniforms.stepSize = value; 62 } 63 }, 64 frustumLength : { 65 get : function() { 66 return generate.uniforms.frustumLength; 67 }, 68 set : function(value) { 69 generate.uniforms.frustumLength = value; 70 } 71 }, 72 randomTexture : { 73 get : function() { 74 return generate.uniforms.randomTexture; 75 }, 76 set : function(value) { 77 generate.uniforms.randomTexture = value; 78 } 79 }, 80 delta : { 81 get : function() { 82 return blur.uniforms.delta; 83 }, 84 set : function(value) { 85 blur.uniforms.delta = value; 86 } 87 }, 88 sigma : { 89 get : function() { 90 return blur.uniforms.sigma; 91 }, 92 set : function(value) { 93 blur.uniforms.sigma = value; 94 } 95 }, 96 blurStepSize : { 97 get : function() { 98 return blur.uniforms.stepSize; 99 }, 100 set : function(value) { 101 blur.uniforms.stepSize = value; 102 } 103 }, 104 ambientOcclusionOnly : { 105 get : function() { 106 return ambientOcclusionModulate.uniforms.ambientOcclusionOnly; 107 }, 108 set : function(value) { 109 ambientOcclusionModulate.uniforms.ambientOcclusionOnly = value; 110 } 111 } 112 }); 113 114 return new PostProcessStageComposite({ 115 name : 'czm_ambient_occlusion', 116 stages : [generateAndBlur, ambientOcclusionModulate], 117 inputPreviousStageTexture : false, 118 uniforms : uniforms 119 }); 120 };
从上面的代码可以看出,代码创建了generate、blur、ambientOcclusionModulate三个处理阶段。generate负责计算屏幕上每个像素的遮挡值,并生成一张灰度图;blur是对generate生成的灰度图就行模糊平滑处理;ambientOcclusionModulate负责根据灰度图对原始场景颜色就行调整。下面分别对这三个阶段进行详细介绍。
2.1 generate代码实现
计算像素遮蔽因子的过程可以概括为:在像素周围,计算采样点对中心像素的遮蔽值,然后对遮蔽值进行累加,最后得到中心像素的遮蔽值。具体实现可以分为两部分:
1:在视空间下计算像素代表面片的法向量
将中心像素以及上下左右的四个像素转换到视空间下,就得到了这五个像素在视空间下的三维位置,记为posInCamera、posInCameraUp、posInCameraDown、posInCameraLeft、posInCameraRight。通过上下左右四个点与中心点的差值得到up、down、left、right四个向量,分别从上下向量、左右向量中选取模较小的向量,记为DX、DY。最后通过normalize(cross(DY, DX))得到中心像素的法向量。整个过程如下图所示:
glsl对应的代码如下:
1 vec4 clipToEye(vec2 uv, float depth) 2 { 3 vec2 xy = vec2((uv.x * 2.0 - 1.0), ((1.0 - uv.y) * 2.0 - 1.0)); 4 vec4 posEC = czm_inverseProjection * vec4(xy, depth, 1.0); 5 posEC = posEC / posEC.w; 6 return posEC; 7 } 8 9 //Reconstruct Normal Without Edge Removation 10 vec3 getNormalXEdge(vec3 posInCamera, float depthU, float depthD, float depthL, float depthR, vec2 pixelSize) 11 { 12 vec4 posInCameraUp = clipToEye(v_textureCoordinates - vec2(0.0, pixelSize.y), depthU); 13 vec4 posInCameraDown = clipToEye(v_textureCoordinates + vec2(0.0, pixelSize.y), depthD); 14 vec4 posInCameraLeft = clipToEye(v_textureCoordinates - vec2(pixelSize.x, 0.0), depthL); 15 vec4 posInCameraRight = clipToEye(v_textureCoordinates + vec2(pixelSize.x, 0.0), depthR); 16 17 vec3 up = posInCamera.xyz - posInCameraUp.xyz; 18 vec3 down = posInCameraDown.xyz - posInCamera.xyz; 19 vec3 left = posInCamera.xyz - posInCameraLeft.xyz; 20 vec3 right = posInCameraRight.xyz - posInCamera.xyz; 21 22 vec3 DX = length(left) < length(right) ? left : right; 23 vec3 DY = length(up) < length(down) ? up : down; 24 25 return normalize(cross(DY, DX)); 26 } 27 28 void main(void) 29 { 30 float depth = czm_readDepth(depthTexture, v_textureCoordinates); 31 vec4 posInCamera = clipToEye(v_textureCoordinates, depth); 32 33 if (posInCamera.z > frustumLength) 34 { 35 gl_FragColor = vec4(1.0); 36 return; 37 } 38 39 vec2 pixelSize = 1.0 / czm_viewport.zw; 40 float depthU = czm_readDepth(depthTexture, v_textureCoordinates- vec2(0.0, pixelSize.y)); 41 float depthD = czm_readDepth(depthTexture, v_textureCoordinates+ vec2(0.0, pixelSize.y)); 42 float depthL = czm_readDepth(depthTexture, v_textureCoordinates- vec2(pixelSize.x, 0.0)); 43 float depthR = czm_readDepth(depthTexture, v_textureCoordinates+ vec2(pixelSize.x, 0.0)); 44 vec3 normalInCamera = getNormalXEdge(posInCamera.xyz, depthU, depthD, depthL, depthR, pixelSize); 45 }
2: 计算周围空间对面片的遮蔽值
在上一步得到了视空间下的面片法向量,接下来就是计算周围空间对面片的遮挡值。过程可以概括为以下几个步骤:
(1)选取要参与遮蔽值计算的空间。周围空间范围的选择是通过在像素坐标系下以中心像素为圆心,以设定值为半径,得到一个圆,但并不是要取圆中的所有像素,因为这样会带来很大的计算量。取而代之的是在四个方向上进行采样,方向值引入一个随机扰动,可以避免出现特别规则的阴影效果。
(2)在选取了方向后,就在该方向上根据设定的采样步长进行采样,得到新的像素,将该像素转换到视空间下,记为stepPosInCamera。
(3)通过stepPosInCamera与posInCamera作差值得到向量diffVec,向量的长度len代表了该点与中心点的距离。通过lengthCap这个值对采样的空间距离进行限制,超出该值的采样点将作废。
(4)通过向量与法向量的点乘,得到值dotVal。dotVal实际表示了两个向量之间夹角的大小,该值越大,表示与法向量夹角越小,遮蔽值越大。
(5)通过len的长度计算该位置遮蔽值的权重,得到一个遮蔽值localAO。
(6)在同一个方向上选取一个最大的localAO值作为该方向上的ao值,然后将四个方向上的ao值相加,并除以4,得到新的ao。
(7)根据给定的intensity参数对ao值就行幂次变换,得到最终的ao值。
选取采样方向的效果如下图所示:
对应的glsl代码如下:
1 float ao = 0.0; 2 vec2 sampleDirection = vec2(1.0, 0.0); 3 float gapAngle = 90.0 * czm_radiansPerDegree; 4 5 // RandomNoise 6 float randomVal = texture2D(randomTexture, v_textureCoordinates).x; 7 8 float inverseViewportWidth = 1.0 / czm_viewport.z; 9 float inverseViewportHeight = 1.0 / czm_viewport.w; 10 11 //Loop for each direction 12 for (int i = 0; i < 4; i++) 13 { 14 float newGapAngle = gapAngle * (float(i) + randomVal); 15 float cosVal = cos(newGapAngle); 16 float sinVal = sin(newGapAngle); 17 18 //Rotate Sampling Direction 19 vec2 rotatedSampleDirection = vec2(cosVal * sampleDirection.x - sinVal * sampleDirection.y, sinVal * sampleDirection.x + cosVal * sampleDirection.y); 20 float localAO = 0.0; 21 float localStepSize = stepSize; 22 23 //Loop for each step 24 for (int j = 0; j < 6; j++) 25 { 26 vec2 directionWithStep = vec2(rotatedSampleDirection.x * localStepSize * inverseViewportWidth, rotatedSampleDirection.y * localStepSize * inverseViewportHeight); 27 vec2 newCoords = directionWithStep + v_textureCoordinates; 28 29 //Exception Handling 30 if(newCoords.x > 1.0 || newCoords.y > 1.0 || newCoords.x < 0.0 || newCoords.y < 0.0) 31 { 32 break; 33 } 34 35 float stepDepthInfo = czm_readDepth(depthTexture, newCoords); 36 vec4 stepPosInCamera = clipToEye(newCoords, stepDepthInfo); 37 vec3 diffVec = stepPosInCamera.xyz - posInCamera.xyz; 38 float len = length(diffVec); 39 40 if (len > lengthCap) 41 { 42 break; 43 } 44 45 float dotVal = clamp(dot(normalInCamera, normalize(diffVec)), 0.0, 1.0 ); 46 float weight = len / lengthCap; 47 weight = 1.0 - weight * weight; 48 49 if (dotVal < bias) 50 { 51 dotVal = 0.0; 52 } 53 54 localAO = max(localAO, dotVal * weight); 55 localStepSize += stepSize; 56 } 57 ao += localAO; 58 } 59 60 ao /= 4.0; 61 ao = 1.0 - clamp(ao, 0.0, 1.0); 62 ao = pow(ao, intensity); 63 gl_FragColor = vec4(vec3(ao), 1.0);
2.2 blur代码实现
对上一步得到的遮蔽值图需要进行平滑,采用的是高斯平滑算法。实现思路就是在水平方向和垂直方向上根据权重叠加相邻像素值,算法讲解可以查询https://developer.nvidia.com/gpugems/GPUGems3/gpugems3_ch40.html。glsl实现代码如下:
1 #define SAMPLES 8 2 3 uniform float delta; 4 uniform float sigma; 5 uniform float direction; // 0.0 for x direction, 1.0 for y direction 6 7 uniform sampler2D colorTexture; 8 9 #ifdef USE_STEP_SIZE 10 uniform float stepSize; 11 #else 12 uniform vec2 step; 13 #endif 14 15 varying vec2 v_textureCoordinates; 16 void main() 17 { 18 vec2 st = v_textureCoordinates; 19 vec2 dir = vec2(1.0 - direction, direction); 20 21 #ifdef USE_STEP_SIZE 22 vec2 step = vec2(stepSize / czm_viewport.zw); 23 #else 24 vec2 step = step; 25 #endif 26 27 vec3 g; 28 g.x = 1.0 / (sqrt(czm_twoPi) * sigma); 29 g.y = exp((-0.5 * delta * delta) / (sigma * sigma)); 30 g.z = g.y * g.y; 31 32 vec4 result = texture2D(colorTexture, st) * g.x; 33 for (int i = 1; i < SAMPLES; ++i) 34 { 35 g.xy *= g.yz; 36 37 vec2 offset = float(i) * dir * step; 38 result += texture2D(colorTexture, st - offset) * g.x; 39 result += texture2D(colorTexture, st + offset) * g.x; 40 } 41 gl_FragColor = result; 42 }
2.3 ambientOcclusionModulate代码实现
ambientOcclusionModulate非常简单,就是根据ao值对原始场景的图片就行明暗处理,产生阴影效果。glsl代码如下:
1 uniform sampler2D colorTexture; 2 uniform sampler2D ambientOcclusionTexture; 3 uniform bool ambientOcclusionOnly; 4 varying vec2 v_textureCoordinates; 5 6 void main(void) 7 { 8 vec3 color = texture2D(colorTexture, v_textureCoordinates).rgb; 9 vec3 ao = texture2D(ambientOcclusionTexture, v_textureCoordinates).rgb; 10 gl_FragColor.rgb = ambientOcclusionOnly ? ao : ao * color; 11 }
3. 总结
Cesium中AO的实现方式属于HBAO(Horizon-based Ambient Occlusion),相对于传统的SSAO,HBAO对阴影的处理效果更好,使场景更加真实。终于把HDAO的实现原理彻底搞明白了,一个字:爽,哈哈哈!!!
PS:Cesium交流可以扫码加群,期待你的加入!!!