以前一直以为OGRE只支持简单的AABB盒射线检测,今天才发现原来OGRE支持精确到三角面的射线检测。
有图有真相,下边上图:

只有当从鼠标位置发出的射线与模型的一个三角面相交时才表示选中该模型,此时点击鼠标就可以选中该模型,使其高亮显示。最左边模型的那一圈黄色表示已经点击鼠标选中该模型,高亮显示以区别其他模型。

上边这张图显示的是选中Y坐标轴,使其颜色改变,表示其被选中,然后在移动模型的时候只改变模型在Y轴的坐标。

源码如下:
bool PickEntity(Ogre::RaySceneQuery* mRaySceneQuery, Ogre::Ray &ray, Ogre::Entity **result, Ogre::uint32 mask ,Ogre::Vector3 &hitpoint, bool excludeInVisible,const Ogre::String& excludeobject, Ogre::Real max_distance)
{
mRaySceneQuery->setRay(ray);
mRaySceneQuery->setQueryMask(mask);
mRaySceneQuery->setQueryTypeMask(Ogre::SceneManager::ENTITY_TYPE_MASK);
mRaySceneQuery->setSortByDistance(true);

if (mRaySceneQuery->execute().size() <= 0) return (false);

// at this point we have raycast to a series of different objects bounding boxes.
// we need to test these different objects to see which is the first polygon hit.
// there are some minor optimizations (distance based) that mean we wont have to
// check all of the objects most of the time, but the worst case scenario is that
// we need to test every triangle of every object.
Ogre::Real closest_distance = max_distance;
Ogre::Vector3 closest_result;
Ogre::RaySceneQueryResult &query_result = mRaySceneQuery->getLastResults();
for (size_t qr_idx = 0; qr_idx < query_result.size(); qr_idx++)
{
// stop checking if we have found a raycast hit that is closer
// than all remaining entities
if ((closest_distance >= 0.0f) && (closest_distance < query_result[qr_idx].distance))
{
break;
}

// only check this result if its a hit against an entity
if ((query_result[qr_idx].movable != NULL) && (query_result[qr_idx].movable->getMovableType().compare("Entity") == 0))
{
// get the entity to check
Ogre::Entity *pentity = static_cast<Ogre::Entity*>(query_result[qr_idx].movable);

if(excludeInVisible)
if (!pentity->getVisible())
continue;
if(pentity->getName() == excludeobject) 
continue;

// mesh data to retrieve
size_t vertex_count;
size_t index_count;
Ogre::Vector3 *vertices;
unsigned long *indices;

// get the mesh information
GetMeshInformationEx(pentity->getMesh(), vertex_count, vertices, index_count, indices,
pentity->getParentNode()->_getDerivedPosition(),
pentity->getParentNode()->_getDerivedOrientation(),
pentity->getParentNode()->_getDerivedScale());
 
//maybe there is a bug in GetMeshInformationEx(),when mesh is a line or a circle, the vertex_count is not multiple of 3
//            if (index_count%3 != 0)
//            {
//                index_count-=index_count%3;
//            }

// test for hitting individual triangles on the mesh
bool new_closest_found = false;
for (int i = 0; i < static_cast<int>(index_count); i += 3) 
{

// check for a hit against this triangle
std::pair<bool, Ogre::Real> hit = Ogre::Math::intersects(ray, vertices[indices[i]],
vertices[indices[i+1]], vertices[indices[i+2]], true, true);

// if it was a hit check if its the closest
if (hit.first)
{
if ((closest_distance < 0.0f) || (hit.second < closest_distance))
{
// this is the closest so far, save it off
closest_distance = hit.second;
new_closest_found = true;
}
}
}

// free the verticies and indicies memory
delete[] vertices;
delete[] indices;

// if we found a new closest raycast for this object, update the
// closest_result before moving on to the next object.
if (new_closest_found)
{
closest_result = ray.getPoint(closest_distance);
(*result) = pentity;
}
}
}

// return the result
if (closest_distance != max_distance)
{
hitpoint = closest_result;
return true;
}
else
{
// raycast failed
return false;
}
}

void GetMeshInformationEx(const Ogre::MeshPtr mesh,
size_t &vertex_count,
Ogre::Vector3* &vertices,
size_t &index_count,
unsigned long* &indices,
const Ogre::Vector3 &position,
const Ogre::Quaternion &orient,
const Ogre::Vector3 &scale)
{
bool added_shared = false;
size_t current_offset = 0;
size_t shared_offset = 0;
size_t next_offset = 0;
size_t index_offset = 0;

vertex_count = index_count = 0;

// Calculate how many vertices and indices we're going to need
for (unsigned short i = 0; i < mesh->getNumSubMeshes(); ++i)
{
Ogre::SubMesh* submesh = mesh->getSubMesh( i );

// We only need to add the shared vertices once
if(submesh->useSharedVertices)
{
if( !added_shared )
{
vertex_count += mesh->sharedVertexData->vertexCount;
added_shared = true;
}
}
else
{
vertex_count += submesh->vertexData->vertexCount;
}

// Add the indices
index_count += submesh->indexData->indexCount;
}


// Allocate space for the vertices and indices
vertices = new Ogre::Vector3[vertex_count];
indices = new unsigned long[index_count];

added_shared = false;

// Run through the submeshes again, adding the data into the arrays
for ( unsigned short i = 0; i < mesh->getNumSubMeshes(); ++i)
{
Ogre::SubMesh* submesh = mesh->getSubMesh(i);

Ogre::VertexData* vertex_data = submesh->useSharedVertices ? mesh->sharedVertexData : submesh->vertexData;

if((!submesh->useSharedVertices)||(submesh->useSharedVertices && !added_shared))
{
if(submesh->useSharedVertices)
{
added_shared = true;
shared_offset = current_offset;
}

const Ogre::VertexElement* posElem = vertex_data->vertexDeclaration->findElementBySemantic(Ogre::VES_POSITION);

Ogre::HardwareVertexBufferSharedPtr vbuf = vertex_data->vertexBufferBinding->getBuffer(posElem->getSource());

unsigned char* vertex = static_cast<unsigned char*>(vbuf->lock(Ogre::HardwareBuffer::HBL_READ_ONLY));

// There is _no_ baseVertexPointerToElement() which takes an Ogre::Real or a double
//  as second argument. So make it float, to avoid trouble when Ogre::Real will
//  be comiled/typedefed as double:
//      Ogre::Real* pReal;
float* pReal;

for( size_t j = 0; j < vertex_data->vertexCount; ++j, vertex += vbuf->getVertexSize())
{
posElem->baseVertexPointerToElement(vertex, &pReal);

Ogre::Vector3 pt(pReal[0], pReal[1], pReal[2]);

vertices[current_offset + j] = (orient * (pt * scale)) + position;
}

vbuf->unlock();
next_offset += vertex_data->vertexCount;
}


Ogre::IndexData* index_data = submesh->indexData;
size_t numTris = index_data->indexCount / 3;
Ogre::HardwareIndexBufferSharedPtr ibuf = index_data->indexBuffer;

bool use32bitindexes = (ibuf->getType() == Ogre::HardwareIndexBuffer::IT_32BIT);

unsigned long*  pLong = static_cast<unsigned long*>(ibuf->lock(Ogre::HardwareBuffer::HBL_READ_ONLY));
unsigned short* pShort = reinterpret_cast<unsigned short*>(pLong);

size_t offset = (submesh->useSharedVertices)? shared_offset : current_offset;

if ( use32bitindexes )
{
for ( size_t k = 0; k < numTris*3; ++k)
{
indices[index_offset++] = pLong[k] + static_cast<unsigned long>(offset);
}
}
else
{
for ( size_t k = 0; k < numTris*3; ++k)
{
indices[index_offset++] = static_cast<unsigned long>(pShort[k]) + static_cast<unsigned long>(offset);
}
}

ibuf->unlock();
current_offset = next_offset;
}
}

使用时用如下代码即可(可能需要根据实际情况有所改动):
Entity* rayResult=NULL;
Vector3 hitPoint;
CEGUI::Point mousePos= CEGUI::MouseCursor::getSingleton().getPosition();
Ray mouseRay = mCamera->getCameraToViewportRay(mousePos.d_x/float(e.state.width),mousePos.d_y/float(e.state.height));
RaySceneQuery* rayQuery=mSceneMgr->createRayQuery(Ray());
if (PickEntity(rayQuery,mouseRay,&rayResult,QF_moveable_object,hitPoint,true))
rayResult->getParentSceneNode(); //find a entity
以上源码参考Ogitor,为了配合自己的程序,稍有改动。
为了方便大家,所以贴出来,建议大家直接看Ogitor源码。

//暂时放着,代码没有测试过!
  • 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