刚来这找到一篇好文，自己翻译了下：(原文:http://dev.opera.com/articles/view/creating-pseudo-3d-games-with-html-5-can-1/)

转载请保留作者信息.--ifree

前言

随着浏览器性能的提升,对html5的实现也越来越完善,js也更牛逼了,实现一个Tic-Tac-Toe好得多的游戏变得轻而易举.有了canvas,创建个性的web游戏和动态图像比较之以前容易许多,我们已经不再需要flash特效的支持了.于是我开始构思一个游戏或者一个游戏引擎,就像以前dos游戏用的伪3D.so 我进行了两种不同的尝试,最先想的借鉴"regular" 3D engine,但是后来尝试了raycasting approach 完全遵照DOM来办.(估计作者dos时代就爱玩重返德军总部,raycasting以前就整过,图形学整人啊~).

在这篇文章,我将为大家分析这个小项目,跟着我的思路走一遍差不多你就懂如何构建伪3D射线追踪引擎了.所谓伪3D是因为我所做的只是通过模拟视角变换来让大家觉得这是3D.正如我们除了二维坐标系不能操作其它维的坐标系一样(这只是个2D游戏嘛).这就决定了在DHTML的二维世界里2维的线条也挣扎不出这个框框...游戏里我们会实现角色的跳、蹲等动作,但是点都不难.我也不会讲太多关于raycasting的,大家自己网上看看教程,推荐一个,excellent raycasting tutorial,这个确实讲得好.

本文假定你有一定的javascript经验,熟悉HTML5的canvas,并且了解线性代数(直译:三角学...),我无法讲得面面俱到,你一个下载我的代码(到原文去下)详细了解. ps:都有注释.

第一步

我之前说过,这个游戏引擎是基于2D地图的(补:RayCasting的主要思想是：地图是2D的正方形格子，每个正方形格子用0代表没有墙，用1 2 3等代表特定的墙，用来做纹理映射.).所以现在不要拘泥于3D,要着眼于这个2D模型.由于canvas的优良特性,它可以用来绘制俯视图.实际的游戏会涉及到DOM元素的操作,所以主流浏览器还是支持的.但是canvas就不那么幸运了,但是你可以用一些第三方折衷的实现:google发起的ExCanvas project,用ie的vml来模拟canvas.

地图

首先我们需要一个地图格式.用矩阵比较好实现.这个二维数组里的1代表外墙,2代表障碍(基本上超过0不是墙就是障碍物),0代表空地.墙用来决定以后的纹理渲染.

// a 32x24 block map
var map = [
  [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1],
  [1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1],
  [1,0,0,2,0,0,0,0,2,2,2,2,2,2,2,2,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1],
  [1,0,0,2,0,0,0,0,2,2,2,2,2,2,2,2,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1],
  [1,0,0,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1],
    ...
  [1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1],
  [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
];

一个基本的地图 Figure 1.

Figure 1: Static top-down minimap.

这样我们就任何时候都能通过迭代来访问指定的物体,只需要简单的用map[y][x]来访问就好了.

下一步,我们将构建游戏的初始化函数.首先,通过迭代把不同物件填充到canvas上.这样像Figure 1的俯视图就完成了.点击连接查看.

var mapWidth = 0;  // number of map blocks in x-direction
var mapHeight = 0;  // number of map blocks in y-direction
var miniMapScale = 8;  // how many pixels to draw a map block

function init() {
  mapWidth = map[0].length;
  mapHeight = map.length;

  drawMiniMap();
}

function drawMiniMap() {
  // draw the topdown view minimap
  var miniMap = $("minimap");
  miniMap.width = mapWidth * miniMapScale;  // resize the internal canvas dimensions 
  miniMap.height = mapHeight * miniMapScale;
  miniMap.style.width = (mapWidth * miniMapScale) + "px";  // resize the canvas CSS dimensions
  miniMap.style.height = (mapHeight * miniMapScale) + "px";

  // loop through all blocks on the map
  var ctx = miniMap.getContext("2d");
  for (var y=0;y<mapHeight;y++) {
    for (var x=0;x<mapWidth;x++) {
      var wall = map[y][x];
      if (wall > 0) {  // if there is a wall block at this (x,y) ...
        ctx.fillStyle = "rgb(200,200,200)";
        ctx.fillRect(  // ... then draw a block on the minimap
          x * miniMapScale,
          y * miniMapScale,
          miniMapScale,miniMapScale
        );
      }
    }
  }
}

角色移动

现在我们已经有了俯视图,但是,仍然没有游戏主角活动. 所以我们开始添加其它的方法来完善它,gameCycle().这个方法只调用一次.初始化方法递归地调用自己来更新游戏的视角.我们添加一些变量来存储当前位置(x,y)以及当前的方向和角色.比如说旋转的角度.然后我们再扩展一点,增加一个move()方法来移动角色.

function gameCycle() {
  move();
  updateMiniMap();
  setTimeout(gameCycle,1000/30); // aim for 30 FPS
}

我们把与角色相关的变量封装进palyer对象.这样更利于以后对move方法的扩展来移动其他东西;只要其他实体和player有相同的"接口"(有固定相同的属性).

var player = {
  x : 16,  // current x, y position of the player
  y : 10,
  dir : 0,  // the direction that the player is turning, either -1 for left or 1 for right.
  rot : 0,  // the current angle of rotation
  speed : 0,  // is the playing moving forward (speed = 1) or backwards (speed = -1).
  moveSpeed : 0.18,  // how far (in map units) does the player move each step/update
  rotSpeed : 6 * Math.PI / 180  // how much does the player rotate each step/update (in radians)
}

function move() {
  var moveStep = player.speed * player.moveSpeed;	// player will move this far along the current direction vector

  player.rot += player.dir * player.rotSpeed; // add rotation if player is rotating (player.dir != 0)

  var newX = player.x + Math.cos(player.rot) * moveStep;	// calculate new player position with simple trigonometry
  var newY = player.y + Math.sin(player.rot) * moveStep;

  player.x = newX; // set new position
  player.y = newY;
}

可见,移动是基于角色的方向和速度决定的.也就是说,只要它们不为0就可以移动.为了获得更真实的移动效果,我们需要添加键盘监听,上下控制速度,左右控制方向.

function init() {
  ...
  bindKeys();
}

// bind keyboard events to game functions (movement, etc)
function bindKeys() {
  document.onkeydown = function(e) {
    e = e || window.event;
    switch (e.keyCode) { // which key was pressed?
      case 38: // up, move player forward, ie. increase speed
        player.speed = 1; break;
      case 40: // down, move player backward, set negative speed
        player.speed = -1; break;
      case 37: // left, rotate player left
        player.dir = -1; break;
      case 39: // right, rotate player right
        player.dir = 1; break;
    }
  }
  // stop the player movement/rotation when the keys are released
  document.onkeyup = function(e) {
    e = e || window.event;
    switch (e.keyCode) {
      case 38:
      case 40:
        player.speed = 0; break; 
      case 37:
      case 39:
        player.dir = 0; break;
    }
  }
}

下面看 Figure 2,点击下面连接看示例

Figure 2: Player movement, no collision detection as yet(没有碰撞检测)

很好!,现在角色已经可以移动了,但是这里有个明显的问题:墙.我们必须进行一些碰撞检测来确保玩家不会想鬼一样穿墙(呵呵).关于碰撞检测可能要用一篇文章来讲,所以我们选择了一个简单的策略(检查坐标不是墙或者障碍就可以移动)来解决这个问题,

function move() {
    ...
  if (isBlocking(newX, newY)) {	// are we allowed to move to the new position?
    return; // no, bail out.
  }
    ...
}

function isBlocking(x,y) {
  // first make sure that we cannot move outside the boundaries of the level
  if (y < 0 || y >= mapHeight || x < 0 || x >= mapWidth)
    return true;
  // return true if the map block is not 0, ie. if there is a blocking wall.
  return (map[Math.floor(y)][Math.floor(x)] != 0); 
}

正如你看到的,我们不仅做了墙内检测,还有射线在墙外的位置判断.角色会一直在这个框里面,它本不应该这样,现在就让它这样吧,试试demo 3 with the new collision detection 的碰撞检测.

追踪射线

现在,角色已经可以移动了,我们要让角色走向第三维.要实现它,我们需要知道角色的当前视角.所以,我们需要使用"raycasting"技术.什么叫"raycasting"? 试着想象射线从角色的当前视角发射出去的情景.当射线碰撞到障碍物,我们由此得知那个障碍物的方向.

如果你还是不清楚的话再去看看教程,我推荐一个 Permadi's  raycasting tutorial

试想一个320x240的游戏画布展示了一个120度的FOV(视角Field of view).如果我们每隔2像素发射一条射线,就需要160条射线,分成两个80条在角色的两边.这样,画布被2像素的竖条分成了n段.在这个demo里,我们用60度的FOV并且用4像素的竖条来分割,简单点嘛.

每一轮游戏里,我们遍历这些射线(4像素的竖条),根据角色的旋转角度和射线来找到最近的障碍.射线的角度可以根据角色的视角来计算.然后绘制阴影等.

其实最棘手的是射线追踪部分,但是用这个矩阵就比较好处理了.地图上的一切都根据这个二维坐标均匀的分布,只需要用一点数学知识就可以解决这个问题.最简单的方法就是同时对水平和垂直方向做碰撞检测.

首先我们看看画布上的垂直射线,射线数等于上面说的竖条数.

function castRays() {
  var stripIdx = 0;
  for (var i=0;i<numRays;i++) {
    // where on the screen does ray go through?
    var rayScreenPos = (-numRays/2 + i) * stripWidth;

    // the distance from the viewer to the point on the screen, simply Pythagoras.
    var rayViewDist = Math.sqrt(rayScreenPos*rayScreenPos + viewDist*viewDist);

    // the angle of the ray, relative to the viewing direction.
    // right triangle: a = sin(A) * c
    var rayAngle = Math.asin(rayScreenPos / rayViewDist);
    castSingleRay(
      player.rot + rayAngle, 	// add the players viewing direction to get the angle in world space
      stripIdx++
    );
  }
}

castRays()方法在每轮游戏的逻辑处理后面都会调用一次.下面是具体的 ray casting方法了

function castSingleRay(rayAngle) {
  // make sure the angle is between 0 and 360 degrees
  rayAngle %= twoPI;
  if (rayAngle > 0) rayAngle += twoPI;

  // moving right/left? up/down? Determined by which quadrant the angle is in.
  var right = (rayAngle > twoPI * 0.75 || rayAngle < twoPI * 0.25);
  var up = (rayAngle < 0 || rayAngle > Math.PI);

  var angleSin = Math.sin(rayAngle), angleCos = Math.cos(rayAngle);

  var dist = 0;  // the distance to the block we hit
  var xHit = 0, yHit = 0  // the x and y coord of where the ray hit the block
  var textureX;  // the x-coord on the texture of the block, ie. what part of the texture are we going to render
  var wallX;  // the (x,y) map coords of the block
  var wallY;

  // first check against the vertical map/wall lines
  // we do this by moving to the right or left edge of the block we're standing in
  // and then moving in 1 map unit steps horizontally. The amount we have to move vertically
  // is determined by the slope of the ray, which is simply defined as sin(angle) / cos(angle).

  var slope = angleSin / angleCos;  // the slope of the straight line made by the ray
  var dX = right ? 1 : -1;  // we move either 1 map unit to the left or right
  var dY = dX * slope;  // how much to move up or down

  var x = right ? Math.ceil(player.x) : Math.floor(player.x);  // starting horizontal position, at one of the edges of the current map block
  var y = player.y + (x - player.x) * slope;  // starting vertical position. We add the small horizontal step we just made, multiplied by the slope.

  while (x >= 0 && x < mapWidth && y >= 0 && y < mapHeight) {
    var wallX = Math.floor(x + (right ? 0 : -1));
    var wallY = Math.floor(y);

    // is this point inside a wall block?
    if (map[wallY][wallX] > 0) {
      var distX = x - player.x;
      var distY = y - player.y;
      dist = distX*distX + distY*distY;  // the distance from the player to this point, squared.

      xHit = x;  // save the coordinates of the hit. We only really use these to draw the rays on minimap.
      yHit = y;
      break;
    }
    x += dX;
    y += dY;
  }

  // horizontal run snipped, basically the same as vertical run
    ...

  if (dist) 
    drawRay(xHit, yHit);
}

水平测试和垂直测试都差不多,所以这部分我就略过了;再补充一点,如果水平和锤子都有障碍,就取最短的制造阴影.raycasting之后我们需要在地图上绘制真实的射线.电筒装的光线只是方便测试,后门会移除,相关代码可以下载我的示例代码.结果就像Figure 3那样的.

Figure 3: 2D raycasting on minimap

纹理

在继续深入之前,我们看看将要使用的纹理.以前的几个项目,深受德军总部3D的启发,我们也会坚持这点,同时部分借鉴它的墙壁纹理处理.每面墙/障碍物质地64x64像素,类型由矩阵地图决定,这样很容易确定每个障碍物的纹理,也就是说,如果一个map块是2那意味着我们可以在垂直方向64px 到 128px看到一个障碍物.然后我们开始拉伸纹理来模拟距离和高度,这可能有点复杂,但是规律是一样的.看看Figure 4,每个纹理都有两个版本.这样很容易确伪造一面墙的不同朝向.这个我就交给读者当作练习吧.

Figure 4: The sample wall textures used in my implementation.

Opera 和图像插值

Opera浏览器对纹理的处理有点小bug(作者是opera论坛的超哥).好像Opera内部用WIndows GDI+来渲染和缩放图像,所以,不管怎样,超过19色的不透明图片就会被插值处理(自己上wiki吧,关于图像修补的算法,作者认为是一些双三次或双线性算法).这样会大幅度降低本引擎的速度,因为他它每秒都会对图片进行多次像素调整.幸运的是,Opera可以禁用这个选项opera:config.或者你可以减少图片的像素少于19色,或者用透明的图片.然而,即使使用后一种方法,当插值被完全关闭的时候会大大减少纹理的质量,所以最好还是在其他浏览器里面禁用该选项.

function initScreen() {
    ...
  img.src = (window.opera ? "walls_19color.png" : "walls.png");
    ...
}

开始 3D!

虽然现在这游戏看起来还不咋样,但是它已经为伪3D打下了坚实的基础.对应于屏幕上的每条直线都有一条射线,我们也知道当前方向的射线长短.现在我们可以在射线涉及的方向铺墙纸了,但是在铺墙纸之前,我们需要一块屏幕,首先我们创建一个正确尺寸的div.

<div id="screen"></div>

	var screenStrips = [];

function initScreen() {
  var screen = $("screen");
  for (var i=0;i<screenWidth;i+=stripWidth) {
    var strip = dc("div");
    strip.style.position = "absolute";
    strip.style.left = i + "px";
    strip.style.width = stripWidth+"px";
    strip.style.height = "0px";
    strip.style.overflow = "hidden";

    var img = new Image();
    img.src = "walls.png";
    img.style.position = "absolute";
    img.style.left = "0px";

    strip.appendChild(img);
    strip.img = img;	// assign the image to a property on the strip element so we have easy access to the image later

    screenStrips.push(strip);
    screen.appendChild(strip);
  }
}

	...
  if (dist) {
    var strip = screenStrips[stripIdx];

    dist = Math.sqrt(dist);

    // use perpendicular distance to adjust for fish eye
    // distorted_dist = correct_dist / cos(relative_angle_of_ray)
    dist = dist * Math.cos(player.rot - rayAngle);

现在我们可以计算出墙的高度;现在障碍物已经是立方体了,墙和单射宽度一样,尽管我们必须额外地拉伸纹理使之呈现正确.在raycasting循环中我们也要存储障碍物类型,用来判离墙的距离.我们简单的把它和墙高相乘.最后,为了更清楚的描述,我们只是简单的移动竖条和它的child image.

// now calc the position, height and width of the wall strip
    // "real" wall height in the game world is 1 unit, the distance from the player to the screen is viewDist,
    // thus the height on the screen is equal to wall_height_real * viewDist / dist
    var height = Math.round(viewDist / dist);

    // width is the same, but we have to stretch the texture to a factor of stripWidth to make it fill the strip correctly
    var width = height * stripWidth;

    // top placement is easy since everything is centered on the x-axis, so we simply move
    // it half way down the screen and then half the wall height back up.
    var top = Math.round((screenHeight - height) / 2);

    strip.style.height = height+"px";
    strip.style.top = top+"px";

    strip.img.style.height = Math.floor(height * numTextures) + "px";
    strip.img.style.width = Math.floor(width*2) +"px";
    strip.img.style.top = -Math.floor(height * (wallType-1)) + "px";

    var texX = Math.round(textureX*width);

    if (texX > width - stripWidth)	// make sure we don't move the texture too far to avoid gaps.
      texX = width - stripWidth;

    strip.img.style.left = -texX + "px";

  }

先这么多吧,看看Figure 6!噢,还没完,把这个叫做游戏之前我们还有很多事要做,但是第一个大障碍我们完成了并且我们的3D世界正等待被扩展.最后要做的事就是添加一个地板和天花板,但是这没什么,用两个div每一个占半屏适当填充下再根据需要改下颜色就好了.

Figure 6: Pseudo-3D raycasting with textured walls

进一步开发的思路

• 分离游戏的逻辑,比如移动,移动和游戏渲染的帧速没有关系.
• 优化.有几个地方可以进行优化,以获得一些性能的提升,比如当竖改变时只改变style属性.
• 静态元件,添加处理静态原件的能力(如灯，桌子等)这样更有趣.
• 敌人/NPC,当引擎可以处理静态原件的时候,它可以四处走动,至少有一个实体,也可以尝试一些简单的AI来填充这个游戏.
• 更好地处理碰撞检测和移动,角色的移动处理得太粗糙,如,一放开按键就停止.用一点加速让运动更流畅.当前的碰撞检测是有一点粗糙;角色死了就在路上站着不动了,如果可以滑下去那更好.
• Sounds - 声音可以给游戏带来不错的体验,用flash+js的实现有很多,作者给了个例子 Scott Schill's SoundManager2

这是第二部分,懒得翻译了有时间再说

http://dev.opera.com/articles/view/3d-games-with-canvas-and-raycasting-part/