19-离线词典生成原理、图像描述子用BoW转化为BoW向量和FeatureVe

ORBSLAM2

---

---

代码

bool Tracking::TrackReferenceKeyFrame()
{
    // Compute Bag of Words vector
    mCurrentFrame.ComputeBoW();//如果词袋没有计算，需要计算计算词袋

ComputeBow（）

void KeyFrame::ComputeBoW()
{
    if(mBowVec.empty() || mFeatVec.empty())//如果词袋是空的
    {
        vector<cv::Mat> vCurrentDesc = Converter::toDescriptorVector(mDescriptors);//计算输入描述子，转化词袋需要的描述子的格式
        // Feature vector associate features with nodes in the 4th level (from leaves up)
        // We assume the vocabulary tree has 6 levels, change the 4 otherwise
        mpORBvocabulary->transform(vCurrentDesc,mBowVec,mFeatVec,4);//词袋模型库进行计算词袋，输入当前的描述子，输出磁带向量以及记录node
    }
}

如何实现tansform

//将一幅图像中所有特征点转化为BoWervector 与 FeatureVector

template<class TDescriptor, class F> //这是一个类模板
void TemplatedVocabulary<TDescriptor,F>::transform(
  const std::vector<TDescriptor>& features,
  BowVector &v, FeatureVector &fv, int levelsup) const
{
  v.clear();//将BoWvector与featureVector给清空
  fv.clear();
  
  if(empty()) // safe for subclasses
  {
    return;
  }
  
  // normalize 
  LNorm norm;
  bool must = m_scoring_object->mustNormalize(norm);//依据评分类型，确定是否需要归一化
  
  typename vector<TDescriptor>::const_iterator fit;
  
  if(m_weighting == TF || m_weighting == TF_IDF)//
  {
    unsigned int i_feature = 0;//遍历图像中的特征点
    for(fit = features.begin(); fit < features.end(); ++fit, ++i_feature)
    {
      WordId id;//叶子节点的word id
      NodeId nid;//FeatureVector中的NodeID
      WordValue w; //叶子节点word对应的权重
      // w is the idf value if TF_IDF, 1 if TF
      // 将当前描述子转化为Word id， Word weight，节点所属的父节点id（这里的父节点不是叶子
//的上一层，它距离叶子深度为levelsup）
      transform(*fit, id, w, &nid, levelsup);//进入
      
      if(w > 0) // not stopped
      { //如果权重大于0
        v.addWeight(id, w);//添加wordID与权重
        fv.addFeature(nid, i_feature);//添加深度
      }
    }
    
    if(!v.empty() && !must)
    {
      // unnecessary when normalizing
      const double nd = v.size();
      for(BowVector::iterator vit = v.begin(); vit != v.end(); vit++) 
        vit->second /= nd;
    }
  
  }
  else // IDF || BINARY
  {
    unsigned int i_feature = 0;
    for(fit = features.begin(); fit < features.end(); ++fit, ++i_feature)
    {
      WordId id;
      NodeId nid;
      WordValue w;
      // w is idf if IDF, or 1 if BINARY
      
      transform(*fit, id, w, &nid, levelsup);
      
      if(w > 0) // not stopped
      {
        v.addIfNotExist(id, w);
        fv.addFeature(nid, i_feature);
      }
    }
  } // if m_weighting == ...
  
  if(must) v.normalize(norm);
}

词袋树

Transform()

template<class TDescriptor, class F>
void TemplatedVocabulary<TDescriptor,F>::transform(const TDescriptor &feature, 
  WordId &word_id, WordValue &weight, NodeId *nid, int levelsup) const
{ //传参：特征描述子、word id 、word权重、记录当前描述子转化为world的所属nodeid以及它距离叶子深度的levelup、距离叶子深度
  // propagate the feature down the tree
  vector<NodeId> nodes;
  typename vector<NodeId>::const_iterator nit;

  // level at which the node must be stored in nid, if given
  const int nid_level = m_L - levelsup;//用总的深度减去定义的深度
  if(nid_level <= 0 && nid != NULL) *nid = 0; // root

  NodeId final_id = 0; // root
  int current_level = 0;//第0层

  do
  {
    ++current_level;//深度+1
    nodes = m_nodes[final_id].children;//列出该节点所有id
    final_id = nodes[0];//将第一个赋值给final_id
 
    double best_d = F::distance(feature, m_nodes[final_id].descriptor);//计算他与传入的描述子的距离

    for(nit = nodes.begin() + 1; nit != nodes.end(); ++nit)
    {//开始迭代该层所有node
      NodeId id = *nit;
      double d = F::distance(feature, m_nodes[id].descriptor);//作比较选出最接近的node
      if(d < best_d)
      {
        best_d = d;
        final_id = id;
      }
    }
    
    if(nid != NULL && current_level == nid_level)//判断当前层数是否和选的层数一致
      *nid = final_id;//是的话记录
    
  } while( !m_nodes[final_id].isLeaf() );//判断是不是叶子节点，不是的话接着循环

  // turn node id into word id
  word_id = m_nodes[final_id].word_id;//返回word的id与权重
  weight = m_nodes[final_id].weight;
}