box2d 基础问答, 有空翻一下


Box2D C++ tutorials - FAQ / gotchas

Last edited: July 14 2013

Box2D FAQ / gotchas


Here are some questions that come up very often that don't really need a whole tutorial topic to discuss them, but still warrant mentioning somewhere to help save time and frustration when others come across them.


Why can't I get bodies to go faster than a certain speed?

There is a maximum movement limit of 2.0 units per time step give in the file b2Settings.h in the source code. This might seem like a really slow speed to have as an upper limit, but this a  distance per time step , not a velocity given in units per second. Units in Box2D are tuned to be meters, so at a typical framerate of 60Hz a body covering 2 units per timestep is moving at 120 m/s or 432 km/h (270 mph), which is not really that slow. You can change the value (b2_maxTranslation) if you need to, just be aware that the values in this file are tuned to work well together so you may get other problems if you change it too much. 
(Note: this is a #define constant used thoughout Box2D so you'll need to recompile the library itself to make the change take effect fully.)

If you found yourself asking this question because you are  using pixels as your units in the physics world , simply changing this setting to a value which allows the movement you want in pixels is probably a bad idea. It's much better to scale all the units for the physics world down by a constant ratio are so that a top speed of 2 units per timestep is more reasonable.


Why is there always a small gap between bodies?

When fixtures overlap they are pushed apart so that they don't overlap, but many cases (particularly with stacked objects) make it difficult to keep this situation stable. The box on top moves down due to gravity and touches the box underneath, then gets pushed back up, and so on, forever. Apart from looking terrible from all the jittering around, this will never stabilize and wastes computing time since the bodies will never be able to sleep. 

To address this problem there is a tiny space around fixtures which is used as a buffer, in which two adjacent fixtures can agree that they are touching, but there is no overlap so no correction needs to be made - they can smile at each other and relax. This is another value in b2Settings, and while you can change it (b2_linearSlop) you should really know what you are doing before you do.
(Remember this is a #define constant used thoughout Box2D so you'll need to recompile the library itself to make the change take effect fully.)
Gotchas The default value is 0.005 or 5 millimeters, and as the description in b2Settings.h reads, it is "chosen to be numerically significant, but visually insignificant". The idea is that if your game world is using units of meters, you will not notice this tiny space because you will be looking at things like people, cars, buildings etc. Oh and crates, yeah. However if for some reason your objects are very small (perhaps you are scaling all the physics units by a constant ratio as in the question above and you scaled the wrong way) then this distance will become noticeable. I would not recommmend changing this one - adjust the scale of objects in the physics world instead so that you are not simulating ant-sized crates.


Why do bodies lose speed after bouncing even when their restitution is set to 1?

Due to the way the correcting impulses are calculated for a collision, a small amount of energy can actually be lost, especially in cases where a body collides with more than one other body at the same time. This means that a perfectly elastic collision is not guaranteed (well, the same goes for real-life situations). In very simple scenarios you may be able to adjust the velocity of the involved bodies after the collision has completed.


I'm applying an upwards force to cancel gravity, but the body still moves slowly downwards...

Check that you are applying the force  before the first time step . If you have the ApplyForce call after the worlds' Step call in your main loop, the body will get one time step at normal gravity and it will have a chance to move down a tiny bit before the gravity is cancelled. Just change the order of the calls in the main loop so that the ApplyForce comes first.


Why do kinematic bodies not respond to impulses or forces?

Unlike most of the Box2D world, kinematic bodies are something we do not have in real life. Although they can move like a dynamic body, they have no mass, which means they have infinite momentum and cannot be pushed or bumped around by anything. Usually the whole point of using a kinematic body is that you don't want it to be affected by for example your player jumping on it, so it wouldn't be much use if it was affected by forces or impulses now would it? Kinematic body movements must be controlled by setting linear and angular velocities directly.


Why do very slow collisions cause no rebound?

To help keep the simulation stable very small bounces are treated as inelastic, regardless of the restitution settings you have given the fixtures. The most common case where this becomes a problem is snooker/billiards games, where even a slow collision with the side walls or the other balls should be calculated accurately. In b2Settings.h you can change the default value (b2_velocityThreshold) for the minimum velocity to consider inelastic. Changing this to a much smaller value should not usually cause any problems.


How can I get the points of fixtures to render them?

You can iterate over the points of a polygon fixture like this:
1
2
3
4
5
6
7
8
9
10
11
12
13
  if ( fixture->GetType() == b2Shape::e_polygon ) //just to be sure
  {
      b2PolygonShape* polygonShape = (b2PolygonShape*)fixture->GetShape();
      int vertexCount = polygonShape->GetVertexCount();
      for (int i = 0; i < vertexCount; ++i)
      {
          //get the vertex in body coordinates
          b2Vec2 bcVertex = polygonShape->GetVertex( i );
          
          //get the vertex in world coordinates
          b2Vec2 wcVertex = fixture->GetBody()->GetWorldPoint( bcVertex );
      }
  }
Most likely you want the world coordinates, to use directly in your rendering routine. I left the body coordinate line in there as an example because if you are using a rendering API like OpenGL, you can apply the body transform before entering the loop, and then just render using the body coordinates - see the  drawing your own objects  topic for an example.


If a ray cast starts inside a fixture, how can I find the exit point?

Ray casts which start inside a fixture will ignore that fixture, so you will need to reverse the points and cast the ray in the other direction to find out where it would leave the fixture.


How can I tell which version of Box2D I'm using?

If you are compiling from source you can check in the file b2Settings.cpp.
If you are using a pre-compiled library you can check the values of:
1
2
3
  b2_version.major
  b2_version.minor
  b2_version.revision


Why does this site mention Box2D version numbers higher than the latest available version?

If you are downloading the Box2D source code .zip file from the  official page , you will have the most recent  major release version . The major release version is usually recommended as a stable version of the source code, but it is not always the newest version in existence. Sometimes development has continued after the last major release and is stored in a 'subversion' repository for easy maintenance. This is available by 'checking out' the source code directly as shown  here  using a subversion client.

Chances are if you are reading this, you may not want to bother with installing a subversion client to get the latest code that way. I will keep a link here to a .zip of the latest source code so you can download it without using subversion. Please note that the latest version can contain experimental or untested features and can change at any time.

Checking latest revision...

You can check the  source code changes  page to see what's happening in the subversion repository.


Why do I get a b2Contact for two fixtures when they're not touching?

You may get this problem if you are looking at the list of contacts for the world with b2World::GetContactList() or for a body with b2Body::GetContactList(). The existence of a contact in these lists does not mean that the two fixtures of the contact are actually touching - it only means their AABBs are touching. If you want to know if the fixtures themselves are touching you can use:
1
  bool reallyTouching = contact->IsTouching();
See the  'anatomy of a collision'  topic for more details.

  • 0
    点赞
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
1 目标检测的定义 目标检测(Object Detection)的任务是找出图像中所有感兴趣的目标(物体),确定它们的类别和位置,是计算机视觉领域的核心问题之一。由于各类物体有不同的外观、形状和姿态,加上成像时光照、遮挡等因素的干扰,目标检测一直是计算机视觉领域最具有挑战性的问题。 目标检测任务可分为两个关键的子任务,目标定位和目标分类。首先检测图像中目标的位置(目标定位),然后给出每个目标的具体类别(目标分类)。输出结果是一个边界框(称为Bounding-box,一般形式为(x1,y1,x2,y2),表示框的左上角坐标和右下角坐标),一个置信度分数(Confidence Score),表示边界框中是否包含检测对象的概率和各个类别的概率(首先得到类别概率,经过Softmax可得到类别标签)。 1.1 Two stage方法 目前主流的基于深度学习的目标检测算法主要分为两类:Two stage和One stage。Two stage方法将目标检测过程分为两个阶段。第一个阶段是 Region Proposal 生成阶段,主要用于生成潜在的目标候选框(Bounding-box proposals)。这个阶段通常使用卷积神经网络(CNN)从输入图像中提取特征,然后通过一些技巧(如选择性搜索)来生成候选框。第二个阶段是分类和位置精修阶段,将第一个阶段生成的候选框输入到另一个 CNN 中进行分类,并根据分类结果对候选框的位置进行微调。Two stage 方法的优点是准确度较高,缺点是速度相对较慢。 常见Tow stage目标检测算法有:R-CNN系列、SPPNet等。 1.2 One stage方法 One stage方法直接利用模型提取特征值,并利用这些特征值进行目标的分类和定位,不需要生成Region Proposal。这种方法的优点是速度快,因为省略了Region Proposal生成的过程。One stage方法的缺点是准确度相对较低,因为它没有对潜在的目标进行预先筛选。 常见的One stage目标检测算法有:YOLO系列、SSD系列和RetinaNet等。 2 常见名词解释 2.1 NMS(Non-Maximum Suppression) 目标检测模型一般会给出目标的多个预测边界框,对成百上千的预测边界框都进行调整肯定是不可行的,需要对这些结果先进行一个大体的挑选。NMS称为非极大值抑制,作用是从众多预测边界框中挑选出最具代表性的结果,这样可以加快算法效率,其主要流程如下: 设定一个置信度分数阈值,将置信度分数小于阈值的直接过滤掉 将剩下框的置信度分数从大到小排序,选中值最大的框 遍历其余的框,如果和当前框的重叠面积(IOU)大于设定的阈值(一般为0.7),就将框删除(超过设定阈值,认为两个框的里面的物体属于同一个类别) 从未处理的框中继续选一个置信度分数最大的,重复上述过程,直至所有框处理完毕 2.2 IoU(Intersection over Union) 定义了两个边界框的重叠度,当预测边界框和真实边界框差异很小时,或重叠度很大时,表示模型产生的预测边界框很准确。边界框A、B的IOU计算公式为: 2.3 mAP(mean Average Precision) mAP即均值平均精度,是评估目标检测模型效果的最重要指标,这个值介于0到1之间,且越大越好。mAP是AP(Average Precision)的平均值,那么首先需要了解AP的概念。想要了解AP的概念,还要首先了解目标检测中Precision和Recall的概念。 首先我们设置置信度阈值(Confidence Threshold)和IoU阈值(一般设置为0.5,也会衡量0.75以及0.9的mAP值): 当一个预测边界框被认为是True Positive(TP)时,需要同时满足下面三个条件: Confidence Score > Confidence Threshold 预测类别匹配真实值(Ground truth)的类别 预测边界框的IoU大于设定的IoU阈值 不满足条件2或条件3,则认为是False Positive(FP)。当对应同一个真值有多个预测结果时,只有最高置信度分数的预测结果被认为是True Positive,其余被认为是False Positive。 Precision和Recall的概念如下图所示: Precision表示TP与预测边界框数量的比值 Recall表示TP与真实边界框数量的比值 改变不同的置信度阈值,可以获得多组Precision和Recall,Recall放X轴,Precision放Y轴,可以画出一个Precision-Recall曲线,简称P-R
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值